IZA DP No. 1805 Redistributive Taxation and Personal Bankruptcy in US States Charles Grant Winfried Koeniger DISCUSSION PAPER SERIES Forschungsinstitut zur Zukunft der Arbeit Institute for the Study of Labor October 2005
IZA DP No. 1805
Redistributive Taxation andPersonal Bankruptcy in US States
Charles GrantWinfried Koeniger
DI
SC
US
SI
ON
PA
PE
R S
ER
IE
S
Forschungsinstitutzur Zukunft der ArbeitInstitute for the Studyof Labor
October 2005
Redistributive Taxation and
Personal Bankruptcy in US States
Charles Grant University of Reading and EUI
Winfried Koeniger
IZA Bonn, University of Bonn and EUI
Discussion Paper No. 1805 October 2005
IZA
P.O. Box 7240 53072 Bonn
Germany
Phone: +49-228-3894-0 Fax: +49-228-3894-180
Email: [email protected]
Any opinions expressed here are those of the author(s) and not those of the institute. Research disseminated by IZA may include views on policy, but the institute itself takes no institutional policy positions. The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center and a place of communication between science, politics and business. IZA is an independent nonprofit company supported by Deutsche Post World Net. The center is associated with the University of Bonn and offers a stimulating research environment through its research networks, research support, and visitors and doctoral programs. IZA engages in (i) original and internationally competitive research in all fields of labor economics, (ii) development of policy concepts, and (iii) dissemination of research results and concepts to the interested public. IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper should account for its provisional character. A revised version may be available directly from the author.
IZA Discussion Paper No. 1805 October 2005
ABSTRACT
Redistributive Taxation and Personal Bankruptcy in US States*
Both personal bankruptcy and redistributive taxes can insure households’ consumption risk and both vary considerably across US states. We derive sufficient conditions under which more redistributive taxation makes bankruptcy exemptions less attractive both for the intra-temporal insurance and for inter-temporal consumption smoothing. Exploiting data variation over time for 18 US states 1980-2003, we find considerable support for our model’s predictions: (i) redistributive taxation and bankruptcy exemptions are negatively correlated; (ii) both policies are associated with more equal consumption growth whereas the effect on unsecured household debt is less clear-cut. JEL Classification: E21, E61, G18 Keywords: personal bankruptcy, consumer credit, redistributive taxes and transfers Corresponding author: Winfried Koeniger IZA P.O. Box 7240 53072 Bonn Germany Email: [email protected]
* This work is part of the Finance and Consumption in the EU Chair research program, sponsored by Findomestic Banca and CETELEM at the European University Institute. We thank Giuseppe Bertola for numerous discussions. Justin Wolfers, Burcu Duygan, Hamish Low, Dan Hamermesh, workshop and conference participants at IZA, the Tinbergen Institute, European University Institute, World Congress of the Econometric Society and ESSLE have provided very helpful suggestions and comments.
1 Introduction
Bankruptcy regulation and redistributive taxation are both important policies in the US. The
average US household receives $1,000 in direct transfers per year (see the authors’ calculation
based on the CPS in table 2). At the same time, roughly 1.5% of US households have filed for
personal bankruptcy in each recent year; in 2003 households defaulted on approximately $120
billion or $1,100 per household each year (see White, forthcoming). Besides the aggregate
importance of both policies, there is substantial variation in the regulation of bankruptcy
and redistributive taxation across US states. For example, bankruptcy exemptions (the
assets that may be kept by the debtor when he defaults on his debt) are generous in Texas
where housing property is fully exempt (subject to an acreage limit) regardless of value but
redistribution through taxes and transfers is less pronounced. In contrast, New York allows
for much smaller bankruptcy exemptions but has a more generous redistribution scheme
through taxes and transfers.
This paper argues that this negative correlation between the two policies can be ex-
plained. If markets are incomplete (for empirical evidence see, for example, Attanasio and
Davis, 1996, or Blundell et al., 2004), both policies help households to insure some of the
income fluctuations that they are not otherwise able to insure. Both personal bankruptcy
and redistributive taxes are attractive for agents in this second-best world but this attraction
is reduced in the presence of the other policy. This is far from obvious since the main motive
for taxes and transfers is to redistribute resources intratemporally whereas the bankruptcy
exemption is crucially associated with intertemporal consumption-smoothing (it is only im-
portant if agents save and borrow across time). Although, in reality, a tax and transfer
system could be used to reallocate resources across time (think of a pension scheme), we
abstract from this in our model. This allows us to contrast the crucial differences between
the two policies in a simple stylized way. Of course, while in principle a tax system could be
devised that replicates the redistribution implied by a bankruptcy law, it is not practically
possible since it would require conditioning taxes and transfers on who would otherwise de-
fault. For this the government would need to know not only the current asset position but
also information on the consumption and saving decision requiring information on discount
rates and expected future income.
Bankruptcy legislation provides a ‘fresh start’ for agents who have been hit by a suf-
ficiently bad shock (see for example Hynes, 2002). Bankruptcy provides insurance since
households receiving bad shocks can default, while households without bad shocks repay at
higher interest rates. We show that redistributive taxes and transfers make this fresh start
1
less attractive as they eliminate some of the ex-post inequality in gross income. Moreover,
redistributive taxation decreases agents’ expected differences in income across time and thus
their desire to borrow. Besides these intuitive and straightforward interactions, the model
describes more subtle ways in which redistributive taxation changes the costs and benefits of
personal bankruptcy. For example, both policies affect the bankruptcy decision and thus the
risk premium over the risk-free interest rate. We also derive intuitive sufficient conditions
under which both policies are substitutes in providing partial insurance. These results help
us to better understand the interaction between both policies at the micro level.
An important contribution of this paper is that we construct new data on bankruptcy
exemptions for 18 US states over a long time period, 1980-2003. We use data from the Con-
sumer Expenditure Survey (CEX) for consumption, the Current Population Survey (CPS)
for income and construct measures for bankruptcy exemptions and redistributive taxation.1
Since all our data have time variation, we can control for state specific unobserved het-
erogeneity. We provide empirical evidence that supports the model’s hypotheses. We find
that the level of the bankruptcy exemption and the extent of redistributive taxation are
negatively correlated. A generous exemption is associated with less redistribution through
taxes and transfers, suggesting that both policies are substitutes. Moreover, to support our
theoretical perspective, we provide more direct evidence that both policies are important
for the smoothing and insurance motive. Both the bankruptcy exemption and redistributive
taxation are associated with less inequality in consumption growth (which directly mea-
sures consumption insurance); whereas empirical evidence for the effect of both policies on
unsecured household debt is less clear-cut.
Of course, we are not the first to analyze bankruptcy or redistributive taxation in the
US. For example, Gropp et al. (1997) and Pavan (2005) investigate the effect of personal
bankruptcy procedures on households’ assets, while Zame (1993) and his references show
theoretically how bankruptcy can provide partial insurance against income fluctuations. In
the context of the recent bankruptcy reforms, Athreya (1999, 2005) and Chaterjee et al.
(2002) calibrate numerical models to gauge how the benefits of bankruptcy compare with
the costs, such as higher interest rates. It is also well-known that redistributive taxation
provides partial insurance if financial markets are incomplete (see the seminal paper of
Varian, 1980, and the empirical evidence in Grant et al., 2003, and their references).
However, to the best of our knowledge, this paper is the first to jointly analyze redis-
1We use the income information in the CPS because more households are surveyed than in the CEX so
that state averages are better measured. Moreover, measurement error in income and in consumption will
be uncorrelated if information is obtained from different surveys.
2
tributive taxation and bankruptcy exemptions, focussing on the intra- and inter-temporal
channels of policy interaction, and to empirically quantify their effects. Most related in this
respect are the analyses of Hansen and Imhoroglu (1992), Bertola and Koeniger (2004) and
Athreya and Simpson (2003). The first two papers analyze interactions between government
redistribution and financial market imperfections. They argue that in a second-best world of
incomplete financial markets, more government redistribution can mitigate the adverse wel-
fare effect of credit constraints. The interactions between redistribution and bankruptcy in
this paper are similar in spirit. But since we allow for bankruptcy, the interactions between
financial market imperfections and redistribution now have effects on both the interest rate
and the level of borrowing. Allowing for bankruptcy has the additional advantage that we
are able to test the predictions of the model with data on US states which vary with respect
to bankruptcy exemptions as well as redistribution through taxes and transfers.
Our simple model allows us to derive some analytical results for a quite general class
of utility functions and probability distributions. We show that parametric assumptions on
the shape of the probability distributions are not innocuous for the sign and size of the
policy interaction. Athreya and Simpson (2003) instead numerically solve a fully dynamic
model to analyze interactions between public insurance and bankruptcy quantitatively. In
their model, market imperfections such as moral hazard play an important role. On the
one hand, bankruptcy might reduce search effort of unemployed agents because it shelters
the consumption of agents especially from long-term shocks. On the other hand, lower
unemployment insurance increases search effort, reduces the unemployment rate, and thus
also lowers default rates. Although problems of hidden action or asymmetric information
are certainly important in the real world, we show in this paper that such imperfections
are not necessary to rationalize why bankruptcy is less attractive if redistribution is more
pronounced, and we believe it is important to understand this simpler environment with fewer
parametric assumptions. Moreover, we do not have the data to exploit a richer modelling
framework in the empirical part. Thus, we rather view our approach as complementary to
the one chosen by Athreya and Simpson.
The rest of the paper is structured as follows. In Section 2 we present a simple model
that analyzes interactions between redistributive taxation and bankruptcy exemption. We
describe the data in Section 3 and discuss the econometric specification in Section 4. We
present our empirical results in Section 5 before we conclude in Section 6.
3
2 A stylized model
We construct a simple model with two periods labelled 1 and 2 in which the bankruptcy
decision is modelled in a standard way (see, for example, White, forthcoming), motivating
our empirical analysis. Its simple structure allows us to derive some analytic results for a
relatively general class of utility functions and probability distributions that illustrate the
interactions between a linear redistributive tax/transfer scheme and a bankruptcy exemption.
Agents are risk-averse and either borrow at interest rate r2 from risk-neutral banks, or
lend at the world risk-free interest rate rf . The interest rate r2 is endogenously determined
and incorporates the bank’s expectation about the agent’s repayment behavior in period 2.
Thus, the interest rate r2 will depend on each agent’s circumstances (we drop the agent-
specific index for convenience). In contrast, the world interest rate rf is exogenous and
constant.2 Since the probability of default is weakly larger than zero, r2 ≥ rf .
Agents are born in period 1 with endowment ω1. We focus on a representative borrower
whose choices are a function of these resources and of expected future endowment draws in
period 2. This focus is justified since the median household owes some unsecured debt in all
US states. Moreover, the choices of savers that hold assets at a risk-free rate rf are rather
uninteresting since they are not directly affected by bankruptcy procedures. Some results
for savers are summarized in Appendix A.2.
Timing Given the endowment ω1 (which we normalize to one without loss of generality),
agents decide how much to borrow and consume in the first period. They know that in period
2 they will receive an uncertain endowment
ω2 = µ + ε2 ,
where µ is known and ε2 is random with mean zero. Agents expect their endowment to grow
in period 2 if µ > 1.
After the endowment draw in period 2, agents decide whether to declare bankruptcy and
how much to consume. Given this setup, total resources in period 2 (before the decision to
declare bankruptcy) are defined as
ρ2 =
{ω2 − (1 + rt)b1 if borrow
ω2 + (1 + rf )a1 if save.
2This is a common assumption in the literature. For example, Athreya (2005) motivates this assump-
tion by noting that the ownership of wealth is fairly concentrated. Thus an exogenous interest rate can
be motivated assuming a small group of agents which holds all assets, and is unaffected by bankruptcy
procedures.
4
Depending on whether the agent has carried positive assets a1 or debt b1 from the previous
period, the total resources are larger or smaller than the current endowment ω2. Total
resources in period 1, ρ1, trivially equal the endowment ω1.
Government policy The government is responsible for bankruptcy law and for taxes
and transfers. Agents are taxed or receive transfers depending on the level of their resources
ρt. We define ρ+ so that agents with resources ρt < ρ+ receive transfers whereas agents
with resources ρt > ρ+ are taxed. To make the model interesting we assume that gov-
ernment redistribution cannot be conditioned on assets or the agent’s consumption/saving
choice. Otherwise the distinction between redistributive taxes and transfers and resources
redistributed because of bankruptcy filings would be arbitrary. In particular, we assume a
tractable linear tax/transfer schedule
τ(ρt − ρ+). (1)
Thus, net resources are defined as
ρt ≡ ρt − τ(ρt − ρ+).
This tax-schedule conveniently summarizes redistribution via the parameter τ , which is
constant over time. A larger τ not only implies a larger marginal tax rate in good states (e.g.
for high draws of εt) but also larger transfers in bad states. Notice that the assets of agents
are taxed, and debt and its interest can be deducted as is realistic in the US for most of
our sample period although tax reforms have implemented some changes (see Makin, 2001).
Moreover, we do not explicitly model the deadweight loss resulting from this policy. We will
discuss this issue further below.
The second policy in the model is the bankruptcy exemption x, the level of resources
that can be kept when the household defaults. We focus on this important variable since
while bankruptcy is regulated at the federal level, states are allowed to set their own level
of exemptions (we describe the bankruptcy law in more detail below).
Bankruptcy decision Agents declare bankruptcy in period 2 if they have borrowed and
their total net resources fall below the exemption level x,
ρ2 < x.
Note that we implicitly assume that the agent first pays taxes and receives transfers before
he makes the bankruptcy decision. This is realistic since US households cannot default on
taxes. The critical level of gross resources below which the agent declares bankruptcy is
ρ∗2 =x− τρ+
1− τ.
5
Not surprisingly, agents with more resources declare bankruptcy if the exemption level x
is higher. In contrast, the effect of τ on ρ∗2 depends on whether agents are net tax payers
or receive transfers at the exemption level x (whether ρ+ is greater than or less than x).
If agents with resources higher than the exemption level receive transfers, if ρ+ > x, then
∂ρ∗2/∂τ < 0. In contrast, ∂ρ∗2/∂τ ≥ 0 if ρ+ ≤ x. For later reference note that the critical
value in terms of endowments is given by
ω∗2 =x− τρ+
1− τ+ (1 + r2)b1. (2)
In our simple model bankruptcy only matters for agents who borrow. Thus we focus on
these agents in our analysis (below we discuss the effect of redistributive taxes on savers).
We analyze the borrower’s problem backwards. We first characterize the effect of taxes τ
and the exemption x on expected utility in the second period for a given level of borrowing
b1.3 This allows us to explore how the two policies interact in providing insurance in the
second period. We then analyze how the level of borrowing in period 1 depends on the two
policies x and τ , for a given interest rate r2. From this we learn how the two policies affect
intertemporal smoothing motives. We are able to provide analytic results for a general class
of utility functions and probability distributions. However, we need to parameterize both
utility and probability in order to characterize the equilibrium and optimal exemption level
in period 1 (for endogenous b1 and r2).
2.1 Intratemporal insurance and policy substitutability
Personal bankruptcy only matters in period 2 if agents have borrowed in period 1. Borrowing
is optimal if the marginal utility in period 1, evaluated at net resources ρ1=ω1, is larger than
the expected marginal utility in period 2 conditional on repayment (evaluated at the net
endowment). That is:
u′(ω1) > β(1 + r2)
∫ ∞
ω∗2
u′(ω2)f(ω2)dω2, (3)
where u(.) is a strictly concave, continuous and differentiable utility function, primes denote
derivatives, ω∗2 is the endowment below which the agent declares bankruptcy, β is the discount
factor, r2 is the interest rate at which the agent can borrow in period 1, and f(.) is the
probability density. Moreover, ω2 = ω2 − τ(ω2 − ρ+) is the net endowment in the second
period if the agent has zero assets (no debt).
3The tax schedule is determined by two parameters in equation (1). We focus on τ , but in Appendix A.3
we show that changing ρ+ has similar effects.
6
The expected utility of a borrower for period 2 is
ub2 =
∫ ∞
ω∗2
u((ω2 − (1 + r2)b1︸ ︷︷ ︸=ρ2
)(1− τ) + τρ+)f(ω2)dω2 (4)
+
∫ ω∗2
ρ∗2
u(x)f(ω2)dω2
+
∫ ρ∗2
−∞u(ω2(1− τ) + τρ+)f(ω2)dω2
where the probability density is assumed such that expected marginal utility remains finite
on the support of the distribution.
The first line of expression (4) contains the utility of a borrowing agent if he repays in
period 2. The second line is the utility if the bankruptcy exemption provides full consumption
insurance. And the third line is the utility if the endowment in period 2 is so low that the
bankruptcy exemption only provides partial insurance. Note that, as is realistic, agents who
default on their debt cannot default on tax payments and can no longer tax deduct their
debt and interest payments.
We illustrate the insurance provided by the bankruptcy exemption in the left picture of
Figure 1 (we discuss the figure further after Remark 2 below). We plot consumption (the
solid line) and net resources (the dashed line) as a function of the gross endowment. The
bankruptcy exemption reduces consumption in good states (in equilibrium, agents pay more
interest when they repay as we derive below) and provides insurance in bad states. Thus,
the bankruptcy exemption facilitates intertemporal consumption smoothing by redistributing
resources across states in the second period. If the gross endowment is in the interval (ρ∗2; ω∗2)
the bankruptcy exemption provides full insurance so that consumption is flat at x. Agents
default partially on their debt. For endowments ω2 < ρ∗2 agents fully default on their debt
(the consumption increase afforded by bankruptcy is largest as measured by the distance
between the solid and dashed line at a given endowment level) but consumption is no longer
constant so that insurance is only partial. That is, although the agent defaults on more
debt in this region, his level of consumption is falling together with his endowment. We
now show how the interest rate r2 of the borrowers is determined and depends on the policy
parameters.
2.1.1 Determination of the interest rate
A risk neutral bank in a competitive banking market sets the interest rate r2 so that it
receives the same expected return as lending on the world market at the risk free rate rf .
7
The arbitrage condition is
∫ ω∗2
ρ∗2+ C1−τ
(ω2 − x− C − τ(ω2 − ρ+))f(ω2)dω2 +
∫ ∞
ω∗2
(1 + r2)b1f(ω2)dω2 = b1(1 + rf ) (5)
where C is the deadweight bankruptcy cost. This cost is borne by the bank, and reflects
deadweight administrative and judicial costs.4 The first integral in the arbitrage equation
is the expected repayment in the states of the world in which the agent partially defaults
whereas the second integral is the expected repayment in the states of the world in which
the agent fully repays.5
Totally differentiating equation (5) using Leibniz rule, we find that for a given level of
borrowing:
dr2
dxb1 =
F (ω∗2)− F (ρ∗2 + C1−τ
) + C1−τ
f(ω∗2)
(1− F (ω∗2)) b1 − Cf(ω∗2)b1
> 0 , (6)
where F (.) is the cumulative distribution function. The derivation is given in Appendix A.3.
The intuition is that a higher exemption level x makes the agent default in more states of
the world (recall equation 2). This increases the interest rate which reflects the higher risk of
default. If there is no deadweight loss, so that C = 0, the size of the effect depends positively
on the ratio of the probability of bankruptcy with partial default, F (ω∗2) − F (ρ∗2), over the
probability of full repayment 1−F (ω∗2). Notice that the interest rate is only affected by the
bankruptcy exemption through those states of nature in which the household repays some,
but not all, its debts. Only for these states can the exemption reduce the repaid amount, and
states with full default are not relevant. For C > 0, dr2/dx|b1 increases since the bankruptcy
cost is borne by the bank. Furthermore, we can show the following:
Remark 1: For a given level of borrowing b1 and negligible bankruptcy cost (C = 0),
a higher tax/transfer τ increases the costliness of the exemption in terms of larger
interest payments:
d(
dr2
dx|b1b1
)
dτ> 0 (7)
4If the cost was borne by the agent, for example as a pure utility cost, the bankruptcy cost would affect
the threshold ω∗2 . In this case, the cost would enter the arbitrage equation (5) only through its effect on
the bounds of the integral but no longer by lowering the payment of the agent. In this case the sufficient
condition of Remark 1 simplifies to the decreasing hazard property without additional restrictions on the
bankruptcy cost.5Note that we implicity assume that the bank does not incur the bankruptcy cost if the agent fully
defaults. This assumption is not essential but is reasonable in our model since it is unclear why the bank
should care to start costly procedures if it knows that it does not receive any net payment.
8
if
f(ρ∗2)1− F (ρ∗2)
>f(ω∗2)
1− F (ω∗2)
∣∣∣∣∂ω∗2/∂τ
∂ρ∗2/∂τ
∣∣∣∣ .
If |∂ω∗2/∂τ | > |∂ρ∗2/∂τ | a necessary condition is that the probability distribution has
decreasing hazard on the interval (ρ∗2; ω∗2). Otherwise decreasing hazard is a sufficient
condition.
Proof: see Appendix A.1.
The decreasing hazard property implies that the expected cost of bankruptcy increases
in terms of larger interest payments (in the states of nature in which the agent repays). If
agents receive transfers at resources smaller or equal than the bankruptcy threshold, the
interval (ρ∗2 ; ω∗2) in which the bankruptcy exemption provides full insurance “shifts to the
left”. With decreasing hazard, this shift makes the interest rate more sensitive to changes
in the exemption level because the relative probability mass associated with bankruptcy
and partial default increases relative to the mass associated with repayment. Recall that in
the interval of endowments with partial default the bank still receives some payment which
decreases as exemptions become more generous. If C > 0, the condition to sign the derivative
in Remark 1 can no longer be interpreted in a straightforward way. As inspection of equation
(6) suggests, for C > 0 the shape of the density also becomes important. Thus, parametric
assumptions on the probability distribution in quantitative models are important for the size
and sign of the policy interaction on the “cost-side”.
2.1.2 Policy interactions
We now turn to the ex ante welfare effect of x and τ for borrowers. Totally differentiating
(4) with respect to the exemption x we find for given b1 that
dub2
dx|b1 = −(1− τ)b1
dr2
dx|b1
∫ ∞
ω∗2
u′((1− τ)ρ2(ω2) + τρ+)f(ω2)dω2 (8)
+ (F (ω∗2)− F (ρ∗2)) u′(x),
where more details on the derivation are in Appendix A.3. The first line of the derivative
captures the cost of the bankruptcy exemption because of higher interest payments in the
good states of the world. This effect is less important if much of the interest payment can be
tax deducted. The second line contains the benefit of a higher exemption in the bad states
9
in which bankruptcy provides full insurance. For C = τ = 0, equation (6) implies that banks
insure agents at an actuarially fair price and the sign of dub2/dx|b1 depends on the sign of
u′(x)−∫∞
ω∗2u′(ρ2(ω2))f(ω2)dω2∫∞
ω∗2f(ω2)dω2
As in White (forthcoming), this expression is positive for risk-averse borrowers with strictly
concave utility since
ρ2(ω2) ≥ x for ω2 ∈ (ω∗2;∞)
and thus
u′(x) > u′(ρ2(ω2)) ∀ ω2 > ω∗2
Thus, for C = 0, full exemption is optimal. Instead for C > 0 and 0 < τ < 1, insurance
is actuarially unfair and the welfare gains from the exemptions are bounded. Nonetheless,
unless bankruptcy costs are prohibitively high, some exemption will improve the welfare of
borrowing agents by reducing consumption fluctuations.
Totally differentiating (4) with respect to the τ we find for given b1 that
dub2
dτ|b1 = −(1− τ)b1
dr2
dτ|b1
∫ ∞
ω∗2
u′((1− τ)ρ2(ω2) + τρ+)f(ω2)dω2 (9)
−∫ ∞
ω∗2
(ρ2 − ρ+)u′((1− τ)ρ2(ω2) + τρ+)f(ω2)dω2
−∫ ρ∗2
−∞(ω2 − ρ+)u′(ω2(1− τ) + τρ+)f(ω2)dω2
The sign of this derivative depends on the parameters τ and ρ+ of the tax/transfer sched-
ule. The expected marginal-utility cost in the good states of the world in which the agent
pays taxes needs to be compared with the expected gains in the bad states in which the
agent receives transfers (the second and third line of the derivative). Finally, redistribution
also affects the bankruptcy decision and thus the interest rate (the first line of the deriva-
tive). For later reference it is important to note that redistributive taxes and transfers lower
consumption inequality and thus ex ante provide some insurance to agents.
We can now investigate whether the welfare gains of the bankruptcy exemption are
smaller if government redistribution already provides more plentiful resources in the bad
states of the world. Differentiating expression (8) with respect to τ , we can show the follow-
ing:
10
Remark 2: For given borrowing b1 and negligible bankruptcy cost (C = 0), redistributive
taxation lowers the welfare gains derived from a higher bankruptcy exemption:
d(
dub2
dx|b1
)
dτ< 0,
if consumers cannot tax-deduct their debt, the probability distribution has increasing
density and satisfies the sufficient condition in Remark 1.
Proof: see Appendix A.1.
The interaction between redistribution and the bankruptcy exemption in period 2 can be
decomposed into five different effects (which correspond to the five lines of the derivative in
the proof in Appendix A.1). The first four effects show how more government redistribution
alters the cost of the bankruptcy exemption whereas the last effect captures changes in the
benefits of the bankruptcy exemption:
(i) Larger transfers increase the cost of the bankruptcy exemption in terms of higher
interest rates in the good states of the world, ω2 ∈ (ω∗2;∞). The sign of the effect follows
from Remark 1.
(ii) A higher marginal tax rate decreases the cost of the bankruptcy exemption in terms of
higher interest rates if consumers can tax deduct interest payments on their debt in the good
states of the world. This makes the bankruptcy exemption more attractive if the marginal
tax rate is higher. In Remark 2 we rule out this policy complementarity as is realistic for
most of the sample period in the empirical part (see Makin, 2001).
(iii) Larger transfers in the bad states of the world imply higher taxes in the good states of
the world when debt is repaid (for ρ2 > ρ+ if ω2 > ω∗2). The higher interest payment resulting
from the bankruptcy exemptions then becomes more costly in marginal-utility terms.
(iv) Larger transfers imply that agents who receive transfers declare bankruptcy only at
a lower endowment ω2 (the integration bounds shift). This increases the probability mass of
states of the world in which the debt is repaid, and thus increases the cost of the exemption
in terms of a larger expected debt burden.
(v) Transfers change the probability mass of the states of the world in which the exemp-
tion fully insures. If
f(ω∗2)− f(ρ∗2) > 0
the probability mass decreases which makes exemption less attractive. Thus, for increasing
density, more redistribution decreases the benefits of bankruptcy exemption.
11
Figure 1 illustrates our findings. Recall that the picture on the left plots consumption
(the solid line) and net resources (the dashed lined) as a function of the gross endowment.
The bankruptcy exemption reduces consumption in good states (because of a higher interest
rate) and provides insurance in bad states. The picture on the right in Figure 1 shows how
redistributive transfers and taxes affect consumption. We have specified parameters such that
the agent receives transfers in all states in which he defaults. This tilts the consumption
function upwards in the bad states of the world and also shifts the bankruptcy threshold
to the left (the consumption function is flat for relatively smaller endowments). Whether
this shift of the consumption function implies that bankruptcy is less desirable cannot be
deduced from the graph immediately. This is because net resources also tilt clockwise and
the interest rate changes. Furthermore, it should be clear that the interaction between the
two policies crucially depends on the probability mass and the changes in marginal utility
attached to the shift of the consumption functions at each level of resources.
In our model we would expect redistributive taxes and the exemptions to be substitutes in
period 2. With concave utility, one would suspect that more consumption in bad states of the
world raises expected utility, but that the marginal increase is lower if more is redistributed
towards these states. However, we have shown that the intratemporal interactions between
the two policies are relatively rich. Redistributive transfers/taxes do not only lower the
marginal benefit of bankruptcy exemptions but also change the threshold at which agents
declare bankruptcy and thus the cost of bankruptcy exemption in terms of higher interest
rates. We have shown that one crucial determinant of the sign of the policy interaction is
the shape of the probability distribution, in particular the importance of increasing density
and decreasing hazard. Such a shape of the probability distribution is of interest because
decreasing hazard and increasing density are properties on the support left of the mode of
realistic and commonly used log-normal distributions. We would expect that bankruptcy
exemptions are relevant especially in that region of the support.
In Remark 2 we have derived comparative statics analyzing τ . A similar exercise could
be performed by exploring changes in ρ+. The insights are similar and are relegated to
Appendix A.3 below.
The model does not allow either policy to affect the other’s deadweight loss. It would be
easy to model this interaction, for example, if we allowed for endogenous labor supply that is
distorted by both policies. However, even with separable preferences over consumption and
leisure, the sign of the interaction then depends on the sign and size of third-order derivatives
of the utility function with respect to consumption and leisure. Thus, we abstract from such
12
interactions for clarity.
We have shown under what conditions intratemporal redistribution in the second period
makes exemption levels granted by bankruptcy laws less attractive. We now analyze how
these policies interact if we allow agents to adjust the amounts borrowed in the first period.
The attractiveness of each policy then depends on the degree of intratemporal inequality
and expected intertemporal inequality. In particular, more borrowing increases the size of
the interval (ρ∗2; ω∗2) in which bankruptcy provides full insurance. Thus, depending on how
redistributive are taxes or how the bankruptcy exemption changes the borrowed amounts,
the exemption may become more or less attractive in period 2.
2.2 Intertemporal smoothing and policy substitutability
The exemption of income in bankruptcy procedures and redistributive taxation do not only
interact in terms of providing insurance in the second period. In our model, exemption lev-
els are more important if resources are quite unequal intertemporally whereas redistribution
through taxation is more effective if inequality is intratemporal. Of course, the interesting
case has both intra- and inter-temporal inequality. In this case, intra-temporal redistribu-
tion can eliminate part of the inter-temporal inequality and thus the need for bankruptcy
regulation. Redistributive taxation then affects the smoothing motive and thus also the
welfare gains derived from the exemption levels. In the next two subsections we analyze this
interaction in more detail. We first derive analytic results conditioning on the interest rate
r2. This is useful in order to better understand the subsequent numerical example in which
we allow the interest rate to adjust.
We want to show that compressing the distribution of net income reduces the desire to
borrow for agents who expect higher gross income in the future. To make this point formally,
we characterize the amount borrowed when the Euler equation
u′(ω1 + b1) = β(1 + r2)
∫ ∞
ε∗2
u′(µ + ε2 − (1 + r2)b1︸ ︷︷ ︸=ρ2
− τ(ρ2 − ρ+))dF (ε2) (10)
is satisfied (implicitly we assume that the parameters are such that agents find it optimal to
borrow because they are impatient enough or anticipate higher future income). Recall that
ω2 = µ + ε2 and that the amount borrowed is only repaid above the bankruptcy threshold
ε∗2 (the effect of borrowing on the margins of the integral in (4) cancel in the derivation of
equation (10)). The bankruptcy threshold is defined as
ε∗2 =x− τρ+
1− τ+ (1 + r2)b1 − µ
13
This threshold depends negatively on expected income in the second period, µ. We show:
Remark 3: For a given interest rate r2
• db1/dx|r2 > 0;
• db1/dτ |r2 < 0 if intertemporal resources are compressed ( ρ+ > ρ1, ρ2 > ρ+) and all
agents with resources less than the exemption level receive transfers ( ρ+ ≥ x).
Proof: see Appendix A.1.
The sign of the derivatives is intuitive. A higher exemption level x insures the agent in
the bad states of the world in period 2: he will repay the debt only for relatively higher
endowment realizations when the cost of repayment in marginal-utility terms is smaller. As
is well known, this makes borrowing more attractive (see, for example White, forthcoming).
Instead, taxation in the good states of the world in which the agent repays increases the
marginal-utility cost of repayment in the second period; and transfers in the first period
lower marginal utility. Both effects make borrowing less attractive. Furthermore, if a larger
τ decreases the bankruptcy threshold ε∗2, for ρ+ > x, debt is repaid in states with higher
marginal utility which makes borrowing less attractive.
Intuitively, if redistribution through taxes and transfers decreases intertemporal inequal-
ity, the desire to borrow falls.6 This lowers the welfare gains derived from the exemption x.
Formally, the interval in which the bankruptcy exemption provides insurance in the second
period depends positively on b1. In the extreme case in which taxes and subsidies completely
align the marginal utility of present consumption with the discounted expected marginal util-
ity of future consumption, agents do not borrow and bankruptcy exemption is useless. We
now provide a numerical example on the policy interaction allowing for borrowing b1 and
the interest rate r2 to be jointly determined.
2.3 Numerical solution
We have shown under what conditions redistributive taxation and the bankruptcy exemp-
tions are substitutes. However, for the derivations on the intratemporal insurance motive
in period 2 we have conditioned on the amount borrowed b1 whereas for the derivations on
the intertemporal smoothing motive in period 1 we have conditioned on the interest rate r2.
With both b1 and r2 endogenous, an interpretable analytic solution is no longer obtainable
6The same holds for saving as shown in Appendix A.2.
14
unless strong assumptions are imposed on the utility function, such as constant absolute risk
aversion. This section numerically illustrates the period 1 equilibrium for constant relative
risk aversion utility:
u(c) =c1−σ − 1
1− σ. (11)
Moreover, we compute the indirect utility of a borrowing agent as a function of the policy
parameters in order to show how the optimal exemption level depends on redistributive taxes
and transfers.
The numerical algorithm to solve for the equilibrium in period 1 is simple: for given
starting values for b1 and r2, we use the Euler equation (10) to iterate for the optimal b1.
We then update the bankruptcy threshold ε∗2 and use the bank’s arbitrage condition (5) to
solve for r2. For the new values of b1 and r2, we restart the algorithm until convergence.7
For illustration purposes we choose an exemption level x of 90% of first-period resources
which is in the range of plausible values for US states (see the data in Tables 4 and 5 and
their discussion below). We assume that ρ+ = 0.9 = x so that the comparative statics for
taxes τ in Remark 3 apply and set the marginal tax rate τ = 0.2 which equals the mean
marginal tax rate for Texas (see Table 5). We assume a rather small bankruptcy cost C of
1.5% in terms of first-period resources. Finally, the coefficient of relative risk aversion σ = 2,
which is well in the range of commonly used values. The parameters are chosen in order
to illustrate the main qualitative insights of our stylized model for borrowing agents rather
than seriously attempting to calibrate a more comprehensive model to US data. With the
chosen parameters, borrowing is optimal (the agent is impatient and expects his income to
rise 40 percent in the second period). Instead the decreasing hazard condition of Remark 1
is certainly violated, since ε2 is assumed to be normally distributed. Nonetheless, the results
of Remark 1 and 2 continue to hold in the numerical example below, making explicit that
this decreasing hazard condition is sufficient but not necessary.
Table 1 summarizes the equilibrium values of interest for the benchmark parametrization
in column (1). Columns (2)-(6) display the results when we change some of the parameters.
Since the first period endowment ω1 = 1, borrowing is expressed as a fraction of it. In
the benchmark case, the agent borrows an amount equivalent to 19% of these resources
7The equilibrium need not always exist. If the exemptions x are so high that the agent is certain to
default, there is no interest rate which clears the market. Moreover, note that in the numerical algorithm we
do not explicitly check whether the government budget constraint is balanced. This would not add any new
insight since the constraint could always be satisfied by making a suitable assumption about the distribution
of initial assets.
15
and defaults on the debt with a probability of 0.014, which is close to empirically observed
frequencies as mentioned in the Introduction. Although the default rate is small, the interest
rate on the debt is 0.032, which is 60% higher than the risk-free rate rf .
In column (2) of Table 1, we increase marginal taxes/transfers to τ = 0.25, approximately
the mean marginal tax rate of US states like Maryland or Minnesota (see Table 5 below).
In our numerical example there are two effects. More redistribution through transfers/taxes
decreases the amount borrowed (as pointed out in Remark 3) and thus the default probability,
but agents are taxed for some endowments where they partially default. Thus the bank
appropriates less in the case of default and overall the interest rate increases slightly in this
example, despite lower frequency of default.
Column (3) shows that a fall in the exemption level to x = 0.85 leaves the amount
borrowed nearly unchanged and does not decrease it as in the comparative statics of Remark
3. The reason is that in our numerical example, the lower exemption level decreases the
probability of default by so much that the interest rate falls substantially. The resulting
wealth, income and substitution effects make borrowing more attractive and cancel the
direct negative effect on borrowing.
In column (4) we lower expected second-period income. Not surprisingly, the direct effect
is that borrowing decreases as mentioned in Remark 3. This effect is strengthened by an
indirect effect: the less plentiful resources in the second period imply a higher probability
of default so that the interest rate rises. In column (5) we increase the cost of bankruptcy
that is borne by the bank to 5% of first-period income. Quite intuitively, this increases the
interest rate charged by the bank, slightly decreases borrowing and also default although the
effects are quantitatively small. Finally, in column (6) we decrease the risk aversion of the
agent. Not surprisingly, this increases the amount borrowed, the probability of default and
thus also the interest rate.
After having characterized the equilibrium numerically, we are interested in computing
the optimal exemption level based on the indirect utility function of the borrower in period 1.
We consider the benchmark case with τ = 0.2 and the case of higher marginal taxes/transfers
τ = 0.25. It turns out that higher marginal taxes/transfers decrease the optimal exemption
level from x∗ = 0.871 to x∗ = 0.845. The implied elasticity is −0.12. This confirms the
conditional results of Remarks 2 and 3 which showed qualitatively how redistribution through
the tax system can lower the welfare gains derived from bankruptcy exemption through an
intra- and intertemporal effect.
16
2.3.1 The role of shock persistence
In models with more than two periods, the persistence of shocks will affect the tradeoff
between the two policies; redistributive taxes and bankruptcy exemptions. In a previous
version of the paper we extended the model to three periods maintaining the assumption
that bankruptcy has no effect on access to the credit market beyond the current period (that
is, there are no dynamic costs of bankruptcy such as exclusion from the credit market in
future periods).8
We found that as shocks become more persistent, bankruptcy becomes more attractive
for a given level of debt and a given interest rate. However, increasing the persistence of
shocks also makes the asset market less useful in smoothing consumption over time (see
Deaton, 1991). Moreover, higher persistence makes banks less willing to lend if they know
that bad shocks persist in the next period and thus negatively affect expected repayment
behavior. Overall, increasing persistence has an ambiguous effect on the optimal exemption
level. We further find that the overall effect of persistence on the optimal exemption level
is mitigated if more pronounced redistribution through taxes and transfers already provides
some insurance.
Assuming dynamic costs of bankruptcy instead, Athreya and Simpson (2003) find a
negative welfare effect of bankruptcy exemptions because exemptions exacerbate the moral
hazard problem of unemployed agents. Athreya and Simpson show in their calibration that
more persistent shocks make bankruptcy less costly. Similarly, calibrating a life-cycle model,
Livshitz et al. (2004) find that bankruptcy decreases welfare if income shocks are transitory
but, in contrast to Athreya and Simpson, they find that bankruptcy increases welfare if
shocks are persistent. Thus, whether and how persistence changes the attractiveness of
bankruptcy is an empirical issue. Unfortunately the available US data did not allow us to
construct a measure for shock persistence that varies across states and over time. Hence we
leave this issue for future research.
We do not explicitly model how the interaction of the two policies analyzed above trans-
late into policy choices. This would be straightforward if we modelled policy choices in a
probabilistic voting model in which agents decide about policies once and for all in period
1 (however, one should be careful with normative conclusions, since we treat the respective
8See Chaterjee et al. (2002) or Livshits et al. (2004) for dynamic models of bankruptcy including such
costs. Empirical evidence by Musto (1999) or Staten (1993) suggests, however, that a substantial fraction
of agents has access to credit in the year immediately after bankruptcy although the stored bankruptcy
information seems to constrain borrowing behavior to some extent.
17
deadweight losses of the two policies as exogenous in our model). If agents with liabilities in
period 1 have the full weight in such a model, all of the insights above immediately translate
into a model of policy choice. This is not unrealistic since the median household in the US
states holds unsecured debt (see the empirical part below for further details).
Ultimately, it is an empirical question whether redistributive taxation and bankruptcy
exemption are substitutes and if so whether their effect on the inter- and intratemporal
insurance motives highlighted by our model can explain why. In particular, we will now
investigate empirically whether both policies matter for consumption insurance and the level
of unsecured household debt.
3 Data
We have provided a model that has described channels of interaction between redistributive
taxation and the level of bankruptcy exemption. We now provide empirical evidence using
interesting variation of the two policies across US states in the period 1980-2003, but first
we discuss the data sources and the econometric specification. Using various data sources,
we construct a sample of state-year cells (except in our debt equations where we use the
household level data). We only mention the most important issues concerning our empirical
application and refer to Appendix B for more detailed information on the data sources.
When constructing year-state cells we restrict the sample to those households for whom
complete state information is available and where the head is aged between 30 to 60. More-
over, farming households are excluded since they have their own bankruptcy regulations. As
frequently done in the literature, we also exclude the self-employed since differences between
business and personal income are hard to distinguish. Furthermore, we are interested in
consumer’s risk and not entrepreneurial risk and self-employed households have motivations
for borrowing other than to smooth consumption. For confidentiality reasons, state informa-
tion is sometimes suppressed in the survey and in some states relatively few households are
sampled. Hence, to ensure that there are enough observations used to construct each cell,
our sample only contains households resident in the 18 largest US states. Appendix B lists
the included states, which cover the full range of US states in terms of geography, taxes and
bankruptcy exemptions.
We use the Consumer Expenditure Survey (CEX) to construct a measure of non-durable
consumption and debt for each household. Our measure of unsecured debt uses separate
questions that were asked about debts held in revolving credit accounts (including store,
18
gasoline, and general purpose credit cards), in installment credit accounts, credit at banks
or savings and loan companies, in credit unions, at finance companies, unpaid medical bills,
and other credit sources. Our measure also includes negative balances held in checking or
brokerage accounts. We sum these different sources of unsecured credit to construct gross
unsecured debt for each household.9 Excluded from the total are mortgage, and other secured
debts so that the debts analyzed differ from those in Gropp et al. (1997). This is important
because the impact of bankruptcy exemptions on secured and unsecured debt ought to be
very different (see Berkowitz and Hynes, 1999). While mortgage (and other secured) debt
is also likely to be important for the household, the creditor has an additional claim to
such assets and can always recover the house (or other security) if the debtor defaults. The
housing, or other exemption will not affect the creditors rights in this case, and hence it
does not make sense to include such debts in the analysis. Consumption and debt have been
deflated by a consumer price index and are in real 1984 dollars. The mean level of debt in
the survey is $2,151 (the median is $331, while the 75th percentile is $2,211). Around 60
percent of people hold at least some unsecured debt, and this proportion is similar across all
states (see the last column in table 5). Thus the median household in each state holds at
least some unsecured debt.
Information on household level income and transfers is obtained from the March sup-
plement of the Current Population Survey (CPS). To measure the level of income taxes
that each household pays, we exploit the TAXSIM 4.0 program developed by Freenberg (see
Freenberg and Coutts, 1993, for details) that is available from the NBER. We construct two
alternative measures of the tax system which account for both federal and state level legisla-
tion on income taxes and transfers: the mean marginal tax rate and an ‘income-compression
measure’ that compares the inequality of gross with net income.10 We prefer the latter mea-
sure because it better accounts for transfers and the different tax rates among agents even
within the same state (see Appendix B for details).
Using legislative data we construct the bankruptcy exemption level for each household
in the CEX sample, which we normalize by dividing the exemption by average income in
each state-year cell. While bankruptcy law is regulated at the federal level, congress allowed
each state to set its own exemption level (those assets, up to some maximum value, that are
9We concentrate on gross rather than net unsecured debt because this is the amount that is not repaid
in bankruptcy.10We also experimented with an inverse poverty index which was defined as one minus the proportion of
households whose after-tax and transfer income is below half the median. The results presented below were
robust to using this alternative measure.
19
exempt from seizure by the creditor when the debtor filed for Chapter 7 bankruptcy). Thus,
state and federal legislation caused exemption levels to differ across household types, across
states, and across time. We exploit this variation to identify the effect of these exemptions
on consumption insurance and unsecured debt holdings.
Table 5, discussed further in Appendix B, shows how much both taxes and the exemptions
vary across states. Taking federal and state taxes together, redistribution through the tax
and transfer system (using the income compression measure) is over a third higher in New
York than in Texas. The table also shows the large differences in the exemption level between
generous states, such as Texas, and stricter states like Maryland. The last column shows
that the median household in each state always holds unsecured debt, which justifies our
focus on borrowers in the theory section.
Our merged data set contains unsecured debt, consumption, consumption growth, bankruptcy
exemption levels and tax redistribution for households in 18 US states during the period 1980-
2003.11 Our debt regressions use the household level data, but for the other regressions we
aggregate the data into 420 state-year cells (we include all states with an average cell-size
of at least 60 - although some states are not identified in the first year or two of the CEX).
This aggregation is necessary to construct measures for consumption inequality.
We construct for each state-year cell both the standard deviation of log consumption and
of consumption growth as a measure for consumption insurance. The standard deviation of
consumption measures the cross-sectional level of consumption inequality. This is a rather
imperfect way to proxy consumption insurance because it does not distinguish between ex-
ante and ex-post inequality. However, Deaton and Paxson (1994) noted that if markets are
complete then the cross-sectional distribution of consumption inequality should not change
over time for a group with fixed membership. They rejected this implication of full insurance
using US data. A useful corollary is that if markets are incomplete then this cross-sectional
measure should increase. Moreover, if the shocks are the same across groups, the rate at
which this inequality changes over time should be larger for those groups in which there is
less risk sharing. While Deaton and Paxson looked at the change in the cross-section of con-
sumption inequality, the same implications arise for the standard deviation of consumption
growth. That is, if markets are complete, consumption should increase for all households
by the same amount. Hence if consumption is growing by different amounts for different
households (in which case the variance of consumption growth is positive), then we can
11For unsecured debt, we only have data from 1988, the first year for which these data are included in the
CEX.
20
again reject complete insurance and moreover, we know that markets provide less insurance
if the variance of consumption growth is larger. Thus, we compute consumption growth
taking the difference between the household’s level of log-consumption in the fifth interview
from the level of log-consumption in the second interview (the household is interviewed five
times in successive quarters in the CEX survey, but no information is released from the first
interview). This measure solves the problem that states may differ in ex-ante within state
consumption inequality.
4 Econometric specification
We estimate three different sets of equations with different dependent variables but otherwise
similar controls. The first regresses unsecured household debt against household characteris-
tics, the bankruptcy exemptions, and, in separate regressions, the two different measures of
the tax system. This investigates the effect of the two policies on the intertemporal smooth-
ing motive. We also regress proxies for consumption insurance against the policy variables.
This quantifies the importance of the two policies in providing consumption insurance. Fi-
nally, we investigate the relationship between the exemption levels and our measures for
tax/transfer redistribution in order to discover whether these two policies are substitutes.
Of course, these specifications do not exploit all the theoretical implications that were
explored, due to the limited number of observations in our data (we only have up to 420
state-year cells and, in some cases, only 60 observations in each cell). However, the reduced-
form regressions allow us to investigate whether the data are broadly consistent with our
theoretical hypotheses.
The consumption insurance regressions on the state-year cells take the form:
yst = β0 + β1τst + β2xst + fs + εst (12)
where as before s is the state, t is the time period, τ is the measure for redistribution
through the tax system, x is the bankruptcy exemption level and yst denotes the dependent
variable. The error is composed of a state fixed effect fs and an idiosyncratic component
εst. Since the dependent variable is the standard deviation of consumption (or consumption
growth), we have removed the between state variation in consumption. This variation would
affect the mean level of consumption in each state, and is removed when constructing the
standard deviation. The state fixed effects capture fixed differences over time in the within
state consumption inequality: they control for all remaining unobserved heterogeneity across
21
states that is not accounted for by the policy variables. For instance, in the regression using
the standard deviation of consumption growth rates, including state fixed effects controls
for differences across states in the variance of permanent income shocks. The state fixed
effects are estimated by including additional state dummies in the regression. Consistent
estimation of β = [β0 β1 β2]′ thus requires a large number of time periods.
We use the CEX to construct our measure of consumption and the CPS for our measure
of the tax system. This has a number of advantages. Since the CPS is a larger survey than
the CEX, cell averages are measured more precisely so that the small sample bias is reduced.
Moreover, if both measures had been taken from the same sample, measurement error of
the dependent variable and the regressor would be correlated. This would not only bias
the estimates but the bias in general would have an ambiguous sign. Constructing the cell
averages using different data sets circumvents this problem.
4.1 Instrumental Variables
Another estimation issue of equation (12) is that the policy variables may change at the same
time as the dependent variables, as is well known in the literature (see for example Besley
and Case, 2000). If a state is hit by a productivity shock, for example, this is likely to affect
the state’s budget (and hence tax requirements) but also gross income of the households
in that state and their level of consumption. Thus, we need to use instruments that can
predict the policy variable τ but do not affect the dependent variable. Instead, we do not
instrument the bankruptcy exemptions for two reasons. Changing the exemptions does not
directly affect the state budget constraint, and changes in the exemptions take longer to
implement.12 Hence it is plausible to treat the bankruptcy exemptions as predetermined in
the regressions.
We experiment with two possible instrument sets: lagged values of the redistribution
measures, and a set of political variables (capturing tastes rather than economic fundamen-
tals) together with measures on the efficiency of the tax system in the state. This instrument
set includes the political affiliation of the state governor and the state legislature, the rela-
tive proportion of voters in each state voting for the democratic rather than the republican
party candidate in the presidential elections and two measures of how effective the state is at
raising tax revenue: the tax fiscal capacity of the state in each period, and the tax intensity
or effort in each period. For the years up to 1991 data on these two variables are available
12Except the automatic updates due to inflation that the federal government and some states implement
at regular intervals.
22
from ACIR (Advisory Commission on Intergovernmental Relations, 1993), while subsequent
data are taken from Tannenwald (2002) and Tannenwald and Turner (2004), although it
was necessary to linearly interpolate the two series for some years. However, these data are
not available for the latest three years in our sample. A full discussion of the variables is
contained in the references mentioned above.
The political variables make useful instruments because they reflect tastes for taxes, re-
distribution. The ACIR measures are even more natural as an instrument as they measure
how efficient the state is at raising tax revenue. States that are more efficient, in the sense
that a given marginal tax rate raises a higher proportion of income from households (ac-
counting for the cost of raising the revenue, and the amount of revenue that is raised) will
have a larger tax-efficiency measure using the ACIR index. Alternatively, to raise a fixed
proportion of income takes less effort by the local tax authorities. A state which is less
efficient at raising tax revenue is more likely to resort to a generous bankruptcy exemption
rather than attempt to increase redistribution through the tax and benefit system.
5 Results
The results are contained in Tables 6 to 8. Table 6 contains regression results on the rela-
tionship between both policies and the level of unsecured household debt. Table 7 displays
results on how both taxes and the exemptions are associated with the standard deviation
of consumption and consumption growth. Finally, Table 8 directly looks at the correlation
between the level of the bankruptcy exemptions and the tax and benefit system.
5.1 Unsecured household debt
Table 6 shows the estimates of the effect of government redistribution through taxes and
transfers and of the bankruptcy exemptions on the level of unsecured debt. In column (1)
we use the income compression measure and in column (2) we use mean marginal tax rates as
a regressor. The regression uses household level data in 1984 prices, with unsecured debt and
the exemption level in logs, or rather as the log(1+b) and log(1+x). We include and report
estimates for a full set of household characteristics which control for observable heterogeneity
that relates to permanent income and life-cycle circumstances and tastes among other things,
and a set of state, year, and month dummies.13
13Since we include an age polynomial and a set of year dummies, this precludes using dummies for year
of birth because they are not linearly independent.
23
Since households cannot report negative debts - such households report zero debts -
estimation must solve this censoring problem. Rather than using a tobit regression, which
imposes that the errors are normally distributed, we estimate the effect of taxes and of
exemptions by a censored least absolute deviation (CLAD) regression as proposed by Powell
(1984).14 This semi-parametric estimator only imposes the weaker assumption that the
error term in the latent regression is symmetrically distributed, and consistently estimates
the median effect.15
The main focus of the analysis is the effect of the bankruptcy exemption and the tax
system. Since we have household level data we can estimate the separate effects of the
exemption on homeowners and on renters. This allows us to at least partially condition on
the assets that the household owns. Hence each regression includes the exemption level,
included separately for homeowners and non-homeowners, and a dummy for the unlimited
homestead exemption for homeowners. This is motivated by Gropp et al. (1997) who showed
that the effect of the bankruptcy exemptions were different for high asset households and
low asset households.16
The estimated effect of redistribution through taxes and transfers is positive but not
significant for both measures of the tax system in columns (1) and (2). This positive sign
is surprising in light of Remark 3 and the numerical exercise: if taxes and transfers reduce
intertemporal inequality, then increasing the amount of redistribution of the tax system
should lower the need to borrow and save to smooth consumption over time. However, the
estimated coefficient is not significant.
The bankruptcy exemption enters negatively for renters and positively for homeowners.
However, the negative effect for renters is small (the coefficient implies that a 10 percent
increase in the exemption reduces debts by 0.3 percent using the income compression mea-
sure, and 0.5 percent using the mean marginal tax rate). These small effects for renters are
neither statistically nor economically significant. The estimated effect on those households
who own their house is significant at the 5 percent level, and the coefficient suggests that
unsecured debts increase by over 2 percent when the bankruptcy exemption increases by 10
percent. These results are qualitatively similar to those obtained by Gropp et al. (1997)
14Results for tobit regressions are qualitatively similar for most regressors and are available on request
from the authors.15The true sample errors of the estimates depend on the unknown density of the errors at the median.
Hence we calculate the standard errors of the estimated coefficients by bootstrapping using 100 draws.16The difference in the effect of the housing and non-housing exemption on debt held by homeowners
was not significant in our estimations. These results, omitted for brevity, justify adding the housing and
non-housing exemptions together.
24
who found that the bankruptcy exemptions had a negative effect on low asset households
and a positive effect on high asset households (although, in contrast to Gropp et al., we only
assign the housing exemption to homeowners in constructing our exemption measure).
The dummy for the unlimited homestead exemption is also positive and significant in the
regression at the 5 percent level. The implied effect is that households have over 50 percent
more debt if they own their own home and live in a state in which the home is fully exempt
from seizure should they declare bankruptcy. However, controlling for state fixed effects, as
in the regression, means that the coefficient of the unlimited homestead exemption dummy
is identified only from states in which this dummy changes over time. This happened only
once when Minnesota abolished the unlimited homestead exemption in 1993 and replaced it
with a homestead exemption of $200,000. Hence, the coefficient of the dummy is not well
identified. Given the poor identification, we will rarely comment on the estimated effect of
this variable in the discussion of the results.
The effect of household characteristics is very similar for both measures of the tax system.
Table 6 reports that younger households (those around 30) have more unsecured debt and
that debts decline steadily with age. This is consistent with standard life-cycle models of
consumer behavior in which income increases over the working life. The table also shows that
better educated households have more debt. This seems reasonable since these households
have higher levels of permanent income relative to current income which they might want
to bring forward at the early stage of their life-cycle. However, households where the head
has completed a full college degree have less unsecured debt than if the head has only had
some college education. Black and female households have lower levels of unsecured debt, as
do married couples. Family size increases the level of debt, but family size squared reduces
it. A similar pattern is apparent for income, and our results show that debts increase with
income over the range of households in our survey. The regression also includes the real
risk-free municipal bond rate as a proxy for the interest rate (we do not observe the interest
rate in the debt contract). The coefficient is negative but statistically insignificant.
5.2 Consumption insurance
Table 7 reports results for consumption inequality for both measures of the tax system,
including state dummies to control for fixed differences across the states. Controlling for state
fixed effects means that identification exploits changes in the policy variables to estimate
their effect on within state inequality. Recall that taking the standard deviation removes the
between group variation. Columns (1) and (5) use the standard deviation of log consumption
25
as the dependent variable. In these regressions, market completeness would imply that
neither taxes nor the bankruptcy exemptions will be significant. Hence if they are significant
we will reject full insurance, as in the Deaton and Paxson test we discussed earlier.
Column (1) uses our preferred measure of the tax system, the income compression mea-
sure. In this regression both the tax system and the bankruptcy exemptions enter negatively
(as predicted by our model in which markets are incomplete), but while the bankruptcy ex-
emption is significant at the 1 percent level, the tax system is not significant. Column (5)
uses the mean marginal tax rate as the measure of the tax system and in this regression
the tax system and the bankruptcy exemption are both significant at the 1 percent level.
In both regressions the coefficient on the unlimited homestead exemption is negative and
significant (although the coefficient is only identified by the 1993 Minnesota reform).
Using the estimates in column (1), the difference between the state with the least and
the state with most income compression explains one tenth of the differences in within state
consumption inequality of the states in our sample (our estimates say nothing about between
state inequality, which would be reflected in differences in the mean of consumption across
states). Instead, taking the difference between the smallest and the largest exemption level
can explain about one quarter of the differences in within state consumption inequality.
The estimated effect is sizeable, but is not implausible even though currently only 1.5
percent of US households formally file for bankruptcy each year. Firstly, we have delib-
erately chosen rather homogeneous groups to construct our sample (we have removed the
self-employed, farmers, households over 60 or under 30, for example) and for the selected
sample the cross-sectional variation in consumption is likely to be smaller than in the general
population. That is, the fraction of explained variation is larger since we remove much of
the variation due to differences in the underlying populations in the different states from the
denominator. Secondly, although only 1.5 percent of households file for bankruptcy, substan-
tially more households default on their debts - and bankruptcy legislation is relevant for these
households since it sets the punishment they would receive if they were pursued. Moreover,
around 23 percent of households in our sample receive public transfers. This means that
many households directly benefit from the redistribution that results from the two policies.
Thirdly, people who do not receive transfers, and do not default, are still affected by the
legislation since they pay higher taxes and pay more interest. Lastly, if prudence matters,
then the consumption behavior of all households is affected by the insurance that the two
policies provide.
As we stated above, the inequality in consumption growth better measures the pure
26
insurance effect of both policies, and this insurance effect is fundamentally what is of interest
to us here. In particular, differences in consumption growth are less contaminated by ex ante
consumption inequality as mentioned above. The rest of the table reports results for this
regression, including state fixed effects. In the OLS regression the tax system is always
significant at the 1 percent level regardless of which measure of the tax system is used.
Moreover, the bankruptcy exemption is significant at the 1 percent level in column (2) and
at 5 percent in column (5). The unlimited homestead exemption, while no longer significant,
remains negative. However, a concern with these regressions is that taxes and income or
consumption shocks may be co-determined. Given the fairly short time frames in which
taxes are decided, they might respond to changes in business conditions in the state. Instead,
bankruptcy exemptions are more difficult to change. Hence we attempt to instrument the
tax redistribution measures using their lags, and using a set of what we call political variables
which are discussed above. In all cases the rank test is passed for conventional significance
levels. Instrumenting the tax measures with their second-order lags gives very similar results.
Again both the tax system and the bankruptcy exemption are significant, and both reduce
the variance of consumption changes, as predicted by our model with incomplete markets.
Lastly, we instrument the tax measures using the political variables we discussed earlier (see
columns 4 and 8). We find that taxes enter the regression significantly, but the bankruptcy
exemptions are no longer significant. Moreover, when we instrument using these variables we
cannot reject the over-identifying restrictions (the Sargan test was passed). In column (4),
the point estimates imply that both policies have an effect of similar size, with the type of
differences across the US states implying around a quarter of a standard deviation reduction
in inequality growth rates when moving from a generous state to an ungenerous state in
each case. Again, while this effect may seem large, we believe it is plausible for the reasons
mentioned above.
These results relate to recent work by Krueger and Perri (2005) who argue that private
provision of credit is enhanced if income is more uncertain; and that households have in-
creased incentives to repay their borrowing since they value access to credit in the future.
In contrast, our paper emphasises that default itself is a useful insurance mechanism in the
presence of uncertainty, notwithstanding possible crowding out of private provision of credit.
This is especially important if income shocks are permanent.
27
5.3 Redistribution and bankruptcy exemption
We have found mixed evidence on how redistribution through taxes and transfers is associ-
ated with unsecured debt but both policies seem to provide some consumption insurance.
This suggests that both redistributive tax-transfer schemes and bankruptcy exemption might
be substitutes according to the hypotheses put forward by the model. In Table 8 we provide
more direct evidence for this hypothesis looking at the correlation between these two policies
in a controlled regression framework. In column (1) of Table 8 we regress the exemption
level on the income-compression measure including a set of state dummies, and a dummy
for Minnesota post 1993 (which limited the housing exemption in that year). We find that
the coefficient of the tax measure is negative and significant at the 1% level when we use
either the income compression measure or the mean marginal tax rate. The coefficient in
column (1) implies a rather small effect: the difference in tax rates between the most and
least generous states would change the exemption level by roughly $100 in 1984 prices.
We also tried instrumenting the tax measures with their lags, and using the set of political
variables which we described earlier. This gives similar results, see columns (2), (3), (4), (6),
(7) and (8). The coefficient is significant at the 1 percent level in all these regression. The
estimated coefficient is 5 times as large when we use the political variables in column (3),
but changes little for the mean marginal tax rates. The regression has been estimated both
with a restricted set of political instruments (including a measure on whether the state
legislature was republican or democrat, and the tax efficiency measure) and with the full
set of instruments discussed above. The restricted set of instruments is a natural subset
since state legislatures must explicitly pass the state budget and the tax efficiency measure
captures the cost of raising a particular amount of taxes. The rank test for the significance
of the instruments in the first-stage regression is passed for both the restricted and full
instrument set. However, the Sargan test rejects the overidentifying restrictions for the full
instrument set and this for both tax measures. For the restricted set of instruments the
Sargan test is passed for our preferred income compression measure of taxes. The implied
elasticity of the exemption with respect to taxes (calculated at the mean) is up to -0.09 when
the income compression measure is used in column (2). This number is of the same order of
magnitude as for the simulation exercise in section 2.3 above.
28
6 Conclusion
We have shown in a simple modelling framework that bankruptcy regulation and redistribu-
tive taxation interact through the intratemporal insurance and intertemporal smoothing
motive. We have provided sufficient conditions under which both policies are substitutes
and search for empirical support using data on US states in the period 1980-2003. Consis-
tent with our theoretical perspective, we have found (i) that both redistributive taxation
and bankruptcy exemptions are negatively correlated with the inequality of consumption
and consumption growth; (ii) that the extent of redistributive taxes and the size of the
bankruptcy exemption level are negatively correlated; and (iii) at least for homeowners, the
exemptions are associated with higher unsecured household debt.
In light of our results, the recent bankruptcy reform passed in the spring of 2005 should
have minor effects on consumption insurance and unsecured debt. The introduced cap of
$125,000 on homestead exemption only applies to property that has been acquired in the
previous 3 years, and the cap is only restrictive in quite generous states like Minnesota
which allows housing property to be exempt up to a value of $200,000. The reform mostly has
changed the bankruptcy procedures which are regulated at the federal level. Households with
above median income face an additional burden of proving that their ‘excess’ current monthly
income (net of ‘required’ expenses) is less than $6,000 or 25 percent of their unsecured debt.
Only in these cases can households file for bankruptcy under chapter 7. However, the rather
vague definitions of excess income and the possibility of households to donate up to 15% of
their income to charity leave substantial room for households to endogenously change their
current level of excess income. Thus, the impact of the reform depends on the enforcement
of the new law by US courts. The additional administrative cost of bankruptcy filings
introduced by the reform might have only mild effects: in our numerical illustration in Table
1, column (5), a change in the deadweight cost C borne by the bank from 1.5 to 5 percent
of current income has only a small effect on the equilibrium.
Our results suggest that the variation redistributive taxation and personal bankruptcy
regulation in the US states can be rationalized within a simple economic model. Although
normative conclusions cannot be drawn with the currently available data, the results of the
regressions with instrumental variables suggest that there is an interesting policy trade-off
in that bankruptcy exemption is less effective in increasing welfare if redistributive taxation
is already pronounced. Such a trade-off is not only relevant for US states but also for many
other developed countries. As surveyed by Tabb (2005) many European countries such
as France and Germany with substantial public welfare programs have recently introduced
29
legislation which allows consumers to declare bankruptcy. Given the tradeoff we investigate,
the additional insurance provided by these reforms is unlikely to be important since welfare
spending is already substantial in these European countries.
30
Appendices
Appendix A: Proofs and extensions
Appendix A.1: Proofs of the Remarks
Proof of Remark 1:
Define ω+2 ≡ ρ∗2 + C/(1− τ). Then
d(
dr2
dx|b1b1
)
dτ=
∂ω∗2∂τ
f(ω∗2)− ∂ω+2
∂τf(ω+
2 ) + C1−τ
∂ω∗2∂τ
f ′(ω∗2) + C(1−τ)2
f(ω∗2)
(1− F (ω∗2)− Cf(ω∗2))
−(F (ω∗2)− F (ω+
2 ) + C1−τ
f(ω∗2)) (−∂ω∗2
∂τf(ω∗2)− C
∂ω∗2∂τ
f ′(ω∗2))
(1− F (ω∗2)− Cf(ω∗2))2 .
The sign of this derivative depends on the numerators which can be rearranged to
∂ω∗2∂τ
f(ω∗2) (1− F (ρ∗2 + C))
−∂ρ∗2∂τ
f(ρ∗2 + C) (1− F (ω∗2))
+ξ(C) ,
where ξ(C) contains all the other terms and ξ(0) = 0. Thus
d(
dr2
dx|b1b1
)
dτ> 0
if
∂ω∗2∂τ
f(ω∗2)1− F (ω∗2)
>∂ρ∗2∂τ
f(ρ∗2)1− F (ρ∗2)
and C = 0. If x < ρ+ implies ∂r2/∂τ < 0, then ∂ω∗2/∂τ < ∂ρ∗2/∂τ < 0 (conditional on b1).
Then a necessary condition for the inequality above to hold is
f(ρ∗2)1− F (ρ∗2)
>f(ω∗2)
1− F (ω∗2).
For ∂r2/∂τ > 0 this inequality is even a sufficient condition. ¥
Proof of Remark 2:
Totally differentiating (8) for given b1,
31
d(
dub2
dx|b1
)
dτ= −d
(dr2
dx|b1b1
)
dτ(1− τ)
∫ ∞
ω∗2
u′(ρ2(ω2))f(ω2)dω2
+b1dr2
dx|b1
∫ ∞
ω∗2
u′(ρ2(ω2))f(ω2)dω2
+dr2
dx|b1b1
∫ ∞
ω∗2
(ρ2(ω2)− ρ+) u′′(ρ2(ω2))f(ω2)dω2
+dr2
dx|b1b1
∂ω∗2∂τ
u′(x)f(ω∗2)
+
(∂ω∗2∂τ
f(ω∗2)−∂ρ∗2∂τ
f(ρ∗2))
u′(x) .
where
∂ω∗2∂τ
<∂ρ∗2∂τ
≤ 0 ,
if the agent receives transfers at the bankruptcy threshold. We assume that redistribution
is such that agents pay taxes if they are able to repay their debt in full: ρ2(ω2)− ρ+ > 0, for
ω2 > ω∗2.
Now recall that the sufficient condition of Remark 1 implies for negligible bankruptcy cost
(C = 0), that d(
dr2
dx|b1b1
)/dτ > 0. If agents cannot tax deduct their debt then the second
line of the derivative vanishes (which otherwise would be positive). Furthermore, given that
dr2/dx|b1 > 0 and utility is strictly concave, u′′(•) < 0, we then know that the third and
fourth line of the derivative are negative. If the density is increasing, f(ω∗2) > f(ρ∗2), then also
the fifth line is negative so that the derivative can be unambiguously signed to be negative.
¥
Proof of Remark 3:
Define net resources in the first period as ρ1 = ρ1 − τ(ρ1 − ρ+). Totally differentiating
the Euler equation (10) we find
db1
dx|r2 = −
β(1+r2)1−τ
u′(x)f(x)
u′′(ρ1 + b1) + β(1 + r2)2(1− τ){∫∞
ε∗2u′′(ρ2)dF (ε2) + u′(x)f(x)
} > 0
db1
dτ|r2 = −
(ρ+ − ρ1)u′′(ρ1 + b1) + β(1 + r2)(ρ2 − ρ+)
∫∞ε∗2
u′′(ρ2)dF (ε2) +
≤0 if x≤ρ+︷︸︸︷∂ε∗2∂τ
u′(x)f(x)
u′′(ρ1 + b1) + β(1 + r2)2(1− τ){∫∞
ε∗2u′′(ρ2)dF (ε2) + u′(x)f(x)
} .
32
The Euler equation implies that b1 is optimally chosen. Thus, the derivative of the Euler
equation with respect to b1 is negative for strictly concave utility functions. Therefore the
denominator of all total derivatives is negative so that db1/dx|r2 > 0.
A larger τ is more likely to decrease b1 if it compresses the resources in period 1 and
period 2: ρ+ > ρ1 and ρ2 > ρ+; and certainly so if ρ+ ≥ x. ¥
Appendix A.2: Algebra for savers
Utility of savers
For agents who decide in period 1 to hold assets a1, expected utility in period 2 is
us2 =
∫ ∞
−∞u([ω2 + (1 + rf )a1] (1− τ) + τρ+)f(ω2)dω2
where a1 are the assets the agent carries from period 1 to period 2 and rf is the exogenous
risk-free world interest rate. Note that the utility of savers only depends on taxes but is
unaffected by the exemption x for given a1.
Lending and redistributive taxation and transfers:
How does τ affect agents who hold positive assets a1? In this case the Euler equation is
u′(ρ1 − a1) = β(1 + rf )
∫ ∞
−∞u′(µ + αε1 + ε2 + (1 + rf )a1︸ ︷︷ ︸
=ρ2
− τ(ρ2 − ρ+))dF (ε2)
It is easy to show that da1/dτ |r < 0 if a larger τ compresses intertemporal resources. For
this to be the case, agents have to receive transfers in period 2 and pay taxes in period 1
(ρ+ > ρ2, ρ1 > ρ+ ; the effect on the bankruptcy threshold obviously is absent here). To
derive this we totally differentiating the Euler equation above to find
da1
dτ|r2 = −(ρ1 − ρ+) u′′(ρ1 − a1)− β(1 + rf )(ρ2 − ρ+)
∫∞−∞ u′′(ρ2)dF (ε2)
u′′(ρ1 + b1) + β(1 + rf )2(1− τ)∫∞−∞ u′′(ρ2)dF (ε2)
.
Appendix A.3: Further derivations of equations
Derivation of dr2/dx|b1Total differentiating equation (5), plugging in ω∗2 and rearranging, we get
((1− F (ω∗2))
(b1 + (1 + r2)
∂b1
∂r2
)− C
∂ω∗2∂r2
f(ω∗2)−∂b1
∂r2
(1 + rf )
)dr2
+
((1− F (ω∗2))
(∂b1
∂x
)−
(F (ω∗2)− F (ρ∗2 +
C
1− τ)
)− C
∂ω∗2∂x
f(ω∗2)−∂b1
∂x(1 + rf )
)dx
= 0.
33
Noting that
∂ω∗2∂x
= 1/(1− τ) + (1 + r2)∂b1
∂x
and
∂ω∗2∂r2
= b1 + (1 + r2)∂b1
∂r2
,
we find the expression dr2/dx|b1 displayed in the text.
Derivation of dub2/dx|b1:
dub2
dx|b1 = −b1
dr2
dx|b1
∫ ∞
ω∗2
u′((1− τ)ρ2(ω2) + τρ+)f(ω2)dω2
+
(∂ω∗2∂x
+∂ρ∗2∂x
)(u(x)− u(x)) f(ω∗2)
+ (F (ω∗2)− F (ρ∗2)) u′(x)
which simplifies to the expression in the text.
Derivation of d(dub
2/dx|b1)/dρ+:
d(
dub2
dx|b1
)
dρ+= −d
(dr2
dx|b1b1
)
dρ+(1− τ)
∫ ∞
ω∗2
u′(ρ2(ω2))f(ω2)dω2
−τdr2
dx|b1b1
∫ ∞
ω∗2
u′′(ρ2(ω2))f(ω2)dω2
− τ
1− τ
dr2
dx|b1b1 u′(x)f(ω∗2)
− τ
1− τ(f(ω∗2)− f(ρ∗2)) u′(x)
It is easy to see that
d(
dr2
dx|b1b1
)
dρ+> 0
under analogous conditions as for dτ . Thus everything is much the same qualitatively but
for the effect in the second line: a larger ρ+ implies less taxation in the good states, thus the
marginal-utility cost of exemption in terms of interest payments decreases. Of course, the
effect of tax-deductible debt is not present here.
34
Appendix B: Data Appendix
The Consumer Expenditure Survey (CEX) is a widely used survey of US households
that has operated on a continuous basis since 1980. It contains detailed information on
both consumer expenditure and the demographic and other characteristics of the household
but it contains less accurate information on saving and secured borrowing. The Bureau of
Labor Statistics (BLS) collects this data to construct the consumer price index and hence
the data-set contains extremely detailed information on the sub-components of consumption
and, crucially, the state of residence of the household. The survey is designed as a rotating
panel, with households being interviewed 5 times at quarterly intervals (although the first is
a contact interview from which no information is made available). Each quarter, households
reaching their fifth interview drop out and are replaced by a new household. As the survey
records detailed information on individual expenditure items, we can construct a measure of
non-durable consumption which includes food and beverages, tobacco, housekeeping services,
fuel, public utilities, repairs, public transport, personal care, entertainment, clothing and
books, each deflated by the appropriate price index. Since 1988, households have also been
asked detailed questions about their financial position, and in particular about all their
outstanding unsecured debt (which includes specific and separate questions on credit-card
debt, debt on store cards, bank debt, debt at savings and loans companies, credit unions,
finance companies and other sources, and medical debt). While each household is interviewed
5 times, questions about unsecured debt are only asked in the second and fifth interview. We
only use information of one of these interviews (to avoid issues concerned with correlation
across observations) and deflate the debt data by average incomes in the state that year
(measured from the CPS). Consumption, on the other hand, was deflated using a household
specific Stone-Geary price index for non-durable expenditure so that it is in 1984 dollars.
Given the detail of the questionnaire, the information on debt in the CEX is of comparable
quality as in the more widely used Survey of Consumer Finances (SCF). Moreover, the
CEX contains state information while the Survey of Consumer Finances only provides state
information in a single year, 1983. In the CEX, the median debt in the whole sample is
$736 for the period 1988-93, and 68% of households hold at least some debt. Conditional
on holding debt, the average debt is $3,984, similar to estimates in Cox and Jappelli (1993)
based on the SCF.
We explained in the main text that only the 18 largest states were included in the analysis
because for the other states the CEX survey has too few observations to sensibly construct
cell averages for consumption or consumption growth. Our sample is limited to states which
35
average at least 60 observations in each cell. The states thus included are: California, Col-
orado, Florida, Georgia, Illinois, Maryland, Massachusetts, Michigan, Minnesota, Missouri,
New York, New Jersey, Ohio, Pennsylvania, Texas, Virginia, Washington and Wisconsin.
Information on household level income and transfers is obtained from the March supple-
ment of the Current Population Survey (CPS). This survey is also managed by the BLS, and
is designed to give very detailed and accurate information on the household’s current income
and demographics. Income is defined as total household labor income. The CPS has the
advantage of being a much larger survey than the CEX. The CPS also contains information
on transfers such as social security and railroad retirement income, supplementary security
income, unemployment compensation, worker’s compensation and veterans payments, public
assistance or welfare, and the value of food stamps received. Some summary statistics of the
annual amount that households receive are reported in Table 2. It shows that 94.3 percent
of the sample households were receiving at least some wage income, and that the average
level of wage income was $34,700 (in current dollars). The average wage among wage-earners
was slightly higher at $36,800. On average the households in the sample received just under
$1,000 of public transfers. Among the 23 percent of households that received at least some
income, this transfer income was much larger, at $4,250. The table also shows some of the
most important components of this transfer income. For example, if agents receive ‘social
security’, average transfers amount to $6,600, although less than four percent of households
receive this category of transfer. The most common transfer are unemployment benefits or
worker compensation, where an average amount of about $2,700 is received by more than
13% of households. Among all households in the sample, the average level of unemployment
benefit was around $350 and this amounts to more than one third of all the received public
transfers.
Measuring redistribution of the tax-transfer system
Constructing a measure of the tax system in each state is not trivial and we need to
address a number of problems. US households are subject to taxes levied at the federal and
state level, by county administrations, and by schoolboards. These taxes include income
taxes, sales taxes, property taxes and duty. We concentrate on income taxes which are
raised at both the federal and state level, but we do not include property and sales taxes in
our measures of the tax system since they are largely levied at the county, schoolboard and
city level which we cannot identify in our data. Moreover, sales taxes are paid at the place of
sale and not that of residence which makes it extremely difficult to devise a measure of sales
taxes levied on each household within a state if cross-border shopping takes place. However,
36
we do not believe that excluding sales taxes is problematic since the expenditures recorded
in the CEX exclude sales taxes, making the consumption measure comparable across states.
Table 3 shows the federal tax schedule in 1998. The marginal tax rate varies from 15% for
single filers whose income is below $26,250 to 39.6% for incomes over $288,350. The income
at which these rates apply are slightly lower for couples filing separately, while if the couple
files jointly the brackets start at twice the income of the couple filing separately. These tax
rates and the tax brackets themselves vary substantially from year to year. Prior to 1996 the
bottom bracket was set at zero, which meant that between 15 and 20% of the low-income
households paid no federal income tax. Furthermore, in 1987, the number of brackets was
considerably reduced.
Taxes also vary considerably across states and over time. As an example, columns two,
three and four of Table 5 display the 1998 tax rates applicable in some of the largest US
states. Over the entire US, eight states, including Texas and Florida in the table, do not
levy any state income tax. The other states have a variety of income tax brackets that
differ in their progressivity. In most states the marginal tax rate increases with income and
in many states there are a variety of tax allowances (which will depend on such things as
whether the taxpayer has a spouse or other dependents). For example, in California, the
lowest tax bracket is at one percent, and the highest is at 9.3 percent. In contrast, other
states like Pennsylvania levy a flat-rate income tax of 2.8 percent and there is no tax-exempt
income. Other states allow some income to be exempt from tax and these tax exemptions
can sometimes be quite large. For example, Minnesota allows the first $2,900 to be exempt
for single filers. A few states, such as California, have a tax credit rather than a level of
exempt income.
To construct the level of income taxes paid by each household, we exploit the TAXSIM
4.0 program. This program uses a variety of household variables, taken from the CPS,
which includes the husband’s and wife’s earnings, interest, dividends and other income, and
information about the household’s characteristics (such as the number of dependent children)
and other deductibles (like property costs) as well as the year and state of residence as inputs.
It uses these variables to calculate both the state and the federal tax bracket, tax liability,
and marginal tax rate for each household in the sample, while explicitly controlling for a
variety of allowances to which the household is entitled.
Having constructed these tax liabilities, the problem is to summarize the tax system in
each state and in each year. One commonly used measure is the mean marginal tax rate
(accounting for both federal and state income taxes). Table 5 shows how mean marginal
37
tax rates vary across some of the largest states. Texas and Florida have the lowest tax rates
of 19 percent since these states have no state income tax and only pay federal income tax.
Tax rates are higher in Maryland and Minnesota, at around 25 percent, reflecting the higher
level of the state-income tax.
However, the mean marginal tax rate is a rather unattractive measure since it does not
allow us to capture the substantial heterogeneity in marginal tax rates across agents each year
even in the same state. For example, a mean marginal tax rate of 20 percent could be due to
a uniform marginal tax rate of 20 per cent, or to the top 20 percent of the population having
a rate of 100 percent and the rest of the population having a rate of zero, or the bottom
20 percent having a 100 percent tax rate and the top 80 percent paying nothing. These
three tax schedules have substantially different implications for redistribution. Moreover,
the mean marginal tax rate ignores the level of transfers that households receive. Thus, we
construct an alternative and more direct measure of how much the tax system redistributes
income, the ‘income-compression measure’:
1− sdst (incomeist − tax liabilityist + transfersist)
sdst(incomeist)
where i denotes the household. This ‘income-compression measure’ compares inequality in
net and gross income for each state s and year t. If inequality in net and gross income are the
same (for example if all households paid the same lump-sum tax), the measure takes the value
of zero. If instead there is no inequality in net income but some inequality in gross income,
the measure takes the value of one. Thus, increasing the amount of redistribution through
taxes and transfers decreases inequality in net income compared with gross income, and
increases the ‘income-compression measure’ of the tax system. Moreover, if all households
faced the same marginal tax rate and there were no allowances, the ‘income-compression
measure’ would be equal to the marginal tax rate (and also the average tax rate). However,
given the substantial heterogeneity in marginal tax rates and transfers across households,
we prefer this ‘income-compression measure’ to the mean marginal tax rate (nonetheless we
will report results for both measures below, as well as for the poverty measure we described
earlier). Table 5 shows that Texas and Florida again have the lowest level of redistribution
using the new measure while the index is now highest in New York, Minnesota and California.
However, the ordering of states is similar across the two measures (the correlation coefficient
is 0.78).
38
Bankruptcy regulation
Bankruptcy in the US is regulated by the Federal Bankruptcy Act of 1978, which contains
two chapters specific to non-farming households. Individuals could choose to file for personal
bankruptcy under either Chapter 7 or under Chapter 13. The act allowed individuals to
discharge their unsecured debts (except alimony, child support, taxes, and student debts)
and make a ‘fresh start’.17
Under chapter 7 of the act, the debtor has all his debts expunged but must surrender all
his assets except those (deemed by the court) necessary for him to make his ‘fresh start’.
These necessary assets are the ‘exemption’, with assets exceeding this value being sold and
the excess amount used to satisfy the debt. Cash (up to the value of the exemption) is
retained by the debtor although in some cases the courts insisted that the money had to be
reinvested in an exempt asset within a certain time period. Under Chapter 13, the debtor
agrees to a repayment schedule for part or all of the debt, but retains his assets. Crucially,
the debtor could choose between chapter 7 or chapter 13, and thus could never be made
to pay more than could be enforced under chapter 7 bankruptcy. Thus the exemptions
under chapter 7 placed an upper limit on the amount of unsecured debt that could be
recovered through the courts by creditors. Our empirical analysis exploits this fact. Around
70 percent of personal bankruptcy cases resulted in a filing for Chapter 7. However, several
courts preferred the debtor to file under chapter 13 but enforced purely nominal repayment
schedules.
The federal exemption levels have been revised several times since 1978 and from April
1998 the bankruptcy act was emended to allow the nominal amounts to be adjusted in line
with the retail price index every three years. The relevant federal exemption levels are shown
in Table 4.18 The 1978 Act allowed the house or homestead to be exempt up to the value of
$7,500 while other exempt assets included a car of $1,200, household goods up to $200 for each
item, jewelry up to $500, other property up to $400 (and any unused homestead exemption),
and ‘tools of trade’ up to $750. This last item refers to work material or assets needed in
order to practise professionally (although some, but not all, jurisdictions allowed transport
to and from work to be included in this category). Throughout the analysis, we will exclude
the ‘tools-of-trade’ exemption since it applies mostly to self-employed households which are
excluded from our sample (although including it in our analysis does not substantively change
17While there was a ‘substantially abuse’ clause in the regulations, households were able to exploit the
provisions of the bankruptcy act without regard to whether the household was genuinely unable to pay or
to whether repayment would result in substantial hardship.18The latest legislation, passed in 2003, lies outside our sample period and hence is not discussed.
39
the results).
Table 4 shows that the 1984 reform introduced an upper limit on the total value of
exempt household goods and reduced the amount of unused homestead exemption that
could be claimed for other goods. The revision in 1994 doubled the dollar amounts in each
category, while from 1998 the dollar amounts have been adjusted for inflation every three
years. Finally, married households who jointly filed for bankruptcy were allowed to claim
the exempt amount in each category for each person.
Bankruptcy had traditionally been regulated by the individual states and the 1978 Act
continued to let states set their own level of exemptions and even to disallow the federal
exemption levels. However, other aspects of the enforcement of bankruptcy law were uniform
across states. Similarly to the federal exemptions, most states have specified a variety of
goods that are exempt from seizure or forced sale and, unless state law prohibited this, the
debtor could choose between the federal and state exemptions (and naturally would choose
the larger of the two exemptions).19
Table 5 displays some of the differences in exemption levels across states for single filers
in 1984 and in 1998. Many states allowed larger exemptions for couples, for older households,
and for households with dependents. The table shows that the homestead was fully exempt
from seizure in Florida and Texas (subject to an acreage limit). Moreover, in Texas in
1998, $30,000 worth of other assets were exempt, with the amount being doubled for couples
filing jointly. In Florida in contrast, the corresponding exemption was up to $1,000 worth of
personal property and a car worth up to $1,000, and households in Florida were not allowed
to claim the federal exemptions. Minnesota allowed the homestead to be fully exempt in 1984
but later changed this to a maximum value of $200,000. The other exemptions increased
from $6,500 to $11,050 during the same period where the exemption level was adjusted in
line with the retail price index every two years. Other states, such as Pennsylvania, set the
exemption level much lower. In Pennsylvania, only $300 of property was exempt from seizure
although clothing was also exempt. However, Pennsylvania allowed households to claim the
federal exemptions and obviously households would prefer to do so in this state. Maryland,
however, both set a low bankruptcy exemption (the housing exemption was $2,500 and
the other exemptions were $3,500) and did not allow the federal exemption to be claimed.
Indeed, Maryland had actually reduced the housing exemption in 1983 from $3,500.
The courts have also allowed debtors substantial room for manoeuvre in fully exploiting
19Information on the state bankruptcy exemptions was constructed from the Annotated State codes and
from primary legislation.
40
all the exemptions available. In many cases courts have allowed the debtor to rearrange his
portfolio of assets prior to default and substitute exempt assets for non-exempt assets (some
limit is placed on the ability to rearrange assets by ‘abuse/fraud’ provisions). Since there is
considerable scope for substituting between assets when filing for bankruptcy, we have added
the exemptions to construct a total nominal value of the exemption for each household in
each state and year. Since households were only allowed to claim the housing exemption
if they owned their own house (either outright or through a mortgage), we have added the
homestead exemption to the exemption on all other assets only if households owned their
home. A more detailed assessment of the household’s asset position is not possible because
of limited information in the CEX. If no specific upper limit of exemption was defined for
a category of goods (for instance Pennsylvania allowed “all necessary wearing apparel”), we
assigned the maximum exemption level for that good category in those jurisdictions that
had a limit (see Grant, 2001 for further details). For the homestead exemption instead,
we include a dummy in the regression if no upper limit for this item was specified. In
calculating the exemption level for each household, we considered the household’s age, the
number of dependents and whether the household was headed by a couple. In states in
which households are allowed to claim the federal exemption and this exemption is higher
than the state exemption, we use the federal exemption to construct our measure. Finally,
we normalize the exemptions by dividing the exemption by average income in each state-
year cell. This normalization will better measure how generous the exemptions are as a
proportion of income.
41
References
[1] Advisory Commission on Intergovernmental Relations (1993): State Fiscal Capacity and
Tax Effort - 1991, Washington, DC: U.S Government Printing Office.
[2] Athreya, Kartik B. (1999): “Welfare Implications of the Bankruptcy Reform Act of
1999,” Journal of Monetary Economics, vol. 49, 1567-1595.
[3] Athreya, Kartik B. (2005): “Fresh Start or Head Start? Uniform Bankruptcy Exemp-
tions and Welfare,” Journal of Economic Dynamics and Control, forthcoming.
[4] Athreya, Kartik B. and Nicole B. Simpson (2003): “Personal Bankruptcy or Public
Insurance?,” Federal Reserve Bank of Richmond Working Paper No. 03-14.
[5] Attanasio, Orazio and Steven J. Davis (1996): “Relative Wage Movements and the
Distribution of Consumption,” Journal of Political Economy, vol. 104, 1227-1262.
[6] Bertola, Giuseppe and Winfried Koeniger (2004): “Consumption Smoothing and the
Structure of Labor and Credit Markets,” IZA Discussion Paper No. 1052.
[7] Berkowitz, Jeremy and Richard M. Hynes (1999): “Bankruptcy Exemptions and the
Market for Mortgage Loans,” Journal of Law and Economics, vol. 42, 809-830.
[8] Besley, Timothy and Anne Case (2000): “Unnatural experiments? Estimating the
incidence of endogenous policies,” Economic Journal, vol. 110, F672-F694.
[9] Blundell, Richard, Luigi Pistaferri and Ian Preston (2004): “Consumption Inequality
and Partial Insurance,” Stanford University, mimeo.
[10] Chaterjee Satyajit, Dean Corbae, Makoto Nakajima and Jose-Vıctor Rıos-Rull (2002):
“A Quantitative Theory of Unsecured Consumer Credit with Risk and Default,” Uni-
versity of Pennsylvania, mimeo.
[11] Cox, Donald and Tullio Jappelli (1993): “The Effect of Borrowing Constraints on Con-
sumer Liabilities,” Journal of Money, Credit and Banking, vol. 25, 197-213.
[12] Deaton, Angus (1991): “Saving and Liquidity Constraints,” Econometrica, vol. 59,
1221-1248.
[13] Deaton, Angus and Christina Paxson (1994): “Intertemporal choice and inequality,”
Journal of Political Economy, vol. 102, 437-467.
42
[14] Freenberg, Daniel and Elisabeth Coutts (1993): “An Introduction to the TAXSIM
model,” Journal of Policy Analysis and Management, vol. 12(1).
[15] Grant, Charles (2001): “Consumer Bankruptcy Law, Credit Constraints and Insurance:
some empirics,” European University Institute, mimeo.
[16] Grant, Charles, Christos Koulovatianos, Alexander Michaelides and Mario Padula
(2003): “Redistributive Policies through Taxation: theory and evidence,” European
University Institute, mimeo.
[17] Gropp, Reint, John K. Scholz and Michelle J. White (1997): “Personal Bankruptcy and
Credit Supply and Demand,” Quarterly Journal of Economics, vol. 112, 217-251.
[18] Hansen, Gary D., and Ayse Imrohoroglu (1992): “The Role of Unemployment Insurance
in an Economy with Liquidity Constraints and Moral Hazard,” Journal of Political
Economy, vol. 100, 118-142.
[19] Hynes, Richard M. (2002): “Optimal Bankruptcy in a Non-Optimal World,” Boston
College Law Review, vol. 44, 1-78.
[20] Krueger, Dirk and Fabrizio Perri (2005): “Does Income Inequality Lead to Consumption
Inequality? Evidence and Theory,” Review of Economic Studies, forthcoming.
[21] Livshitz, Igor, James McGee and Michele Tertilt (2004): “Consumer Bankruptcy: a
fresh start,” Stanford University, mimeo.
[22] Makin, Dean M. (2001): “Household Debt and the Tax Reform Act of 1986,” American
Economic Review, vol. 91, 305-319.
[23] Musto, David K. (1999): “The Reacquisition of Credit Following Chapter 7 Personal
Bankruptcy,” Wharton School, University of Pennsylvania, Working Paper 99-22.
[24] Pavan, Marina (2005): “Consumer Durables and Risky Borrowing: the effect of
bankruptcy protection,” Boston College, mimeo.
[25] Powell, James L. (1984): “Least Absolute Deviations for the Censored Regression
Model,” Journal of Econometrics, vol. 25, 303-325.
[26] Staten, Michael E. (1993): “The Impact of Post-Bankruptcy Credit on Personal
Bankruptcies,” Credit Research Center, Working Paper No. 58, Purdue University.
43
[27] Tabb, Charles J. (2005): “Lessons from the Globalization of Consumer Bankruptcy,”
Illinois Law and Economics Working Paper Series No. LE05-013, University of Illinois,
College of Law.
[28] Tannenwald, Robert (2002): “Interstate Fiscal Disparity in 1997,” New England Eco-
nomic Review, 17-33.
[29] Tannenwald, Robert and Nicholas Turner (2004): “Interstate Fiscal Disparity in State
Fiscal Year 1999,” Public Policy Discussion Papers, Federal Reserve Bank of Boston
No. 04-9.
[30] Varian, Hal (1980): “Redistributive Taxation as Social Insurance,” Journal of Public
Economics, vol. 14, 49-68.
[31] White, Michelle (forthcoming): “Bankruptcy and Consumer Behavior: Theory and US
Evidence,” in: The Economics of Consumer Credit, Bertola, Giuseppe, Richard Disney,
and Charles Grant, eds., ch. 7, MIT Press, Cambridge.
[32] Zame, William R. (1993): “Efficiency and the Role of Default when Security Markets
are Incomplete,” American Economic Review, vol. 83, 1142-1164.
44
Figure 1: Consumption as a function of gross endowment for positive exemption x and
different transfers τ
Table 1: Equilibrium values of borrowing b1, interest rate r2, and default probabilityVariables Benchmark τ = 0.25 x = 0.85 µ = 1.3 C = 0.05 σ = 1
(1) (2) (3) (4) (5) (6)borrowing b1 0.1854 0.179 0.186 0.139 0.1846 0.21interest rate r2 0.0321 0.0325 0.025 0.045 0.034 0.037default prob. 0.0137 0.012 0.005 0.025 0.0136 0.021
In the benchmark case, the risk-free interest rate rf = 0.02, ω1 = 1, the bankruptcy cost C = 0.015, the
bankruptcy exemption x = 0.9, the coefficient of risk-aversion σ = 2, the discount rate β = (1 + 0.1)−1,
µ = 1.4, the marginal tax rate τ = 0.2 while the temporary income shock ε ∼ N(0, 0.1 ∗ ω2)
Table 2: The level of wages and transfers for households in the US
average average if received % receive
wages 34,696 36,789 94.3social security 261 6,601 3.9supplementary security income 77 4,161 1.8unemployment/workers compensation 353 2,688 13.1public assistance / welfare 176 3,712 4.7food stamps 128 1,571 8.1
total transfer 997 4,250 23.4
Data are constructed from reported responses in the March supplement of the CPS for the
years 1980-2003. Total transfer refers to the sum of social security benefits, supplementary
security benefits, unemployment or workers compensation, welfare or other public assistance,
and food stamps. The CPS questionnaire conflates social security benefits with railroad
retirement income, and worker’s compensation with veterans payments.
Table 3: Income thresholds for 1998 federal tax brackets
Tax Rate Tax Bracket
(%) single married jointly married separately % paying
15 0 0 0 58.228 26,250 43,850 21,925 34.231 63,550 105,950 52,975 5.236 132,660 161,450 80,725 1.8
39.6 288,350 288,350 144,175 0.3
The data were made available by the Federation of Tax Administrators at 444 N. Capital Street, Wash-
ington DC. In the table ‘single’ refers to single filers, ‘married jointly’ refers to married couples filing
jointly, while ‘married separately’ refers to married couples who file separate tax returns. ‘% paying’
refers to the proportion of households in the tax bracket. The amounts for the tax bracket refer to the
income at which the tax bracket starts.
46
Table 4: Chapter 7 Exemptions under the Federal Bankruptcy Act
Description Amount Comments$
1978 Exemptions:1. House 7,5002. Car 1,2003. Household Goods no limit on aggregate amount that
can be claimed under this category.4. Jewelry 500 personal use only.5. Other Property Allowed all of unclaimed exemption
from (1).6. Tools of Trade 750 Items needed for job.Revised Exemptions of 1984:3. Household Goods 4,000 $200 each item. (furnishings, goods,
clothes, appliances, books, animals,musical instruments) for personaluse only.
5. Other Property 400 + $3,750 of (1) that is unused.Revised Exemptions of 1994:1. House 15,0002. Car 2,4003. Household Goods 8,000 $400 each item.4. Jewelry 1,0005. Other Property 800 + $7,500 of (1) that is unused.6. Tools of Trade 1,500Revised Exemptions of 1998:1. House 16,1502. Car 2,5753. Household Goods 8,625 $425 each item.4. Jewelry 1,0755. Other Property 850 + $8,075 of (1) that is unused.6. Tools of Trade 1,625Revised Exemptions of 2001:1. House 17,4252. Car 2,7753. Household Goods 9,300 $450 each item.4. Jewelry 1,150 personal use only.5. Other Property 925 + $8,725 of (1) that is unused.6. Tools of Trade 1,750
Source: Title, 11, Section 522(d) of the annotated federal code. Section 104 specified that the amounts
were to be updated with the inflation rate every 3 years, commencing on April 1st 1998. While
not recorded, the federal legislation also allowed (with some limits) insurance policies, pensions and
annuities, social security payments, and awards adjudicated by the courts to be exempted.
47
Tab
le5:
Tax
redis
trib
uti
onan
dban
kru
ptc
yex
empti
ons
by
stat
e
Sta
teTax
esB
ankru
ptc
yE
xem
pti
ons
min
.m
ax.
exem
pt
mar
ginal
inco
me
hou
seot
her
hou
seot
her
fed
deb
tors
bra
cket
bra
cket
rate
com
pre
ssio
n‘8
4‘8
4‘9
8‘9
8
Cal
ifor
nia
1.0
9.3
7222
.834
.330
,000
5,20
050
,000
10,9
0019
8459
.7Flo
rida
no
stat
ein
com
eta
x19
.227
.0no
lim
it1,
000
no
lim
it2,
000
1979
55.5
Mar
yla
nd
2.0
4.75
1,85
025
.132
.62,
500
3,50
02,
500
3,50
019
8262
.9M
innes
ota
5.35
7.85
2,90
024
.634
.3no
lim
it6,
500
200,
000
11,0
5069
.4N
ewY
ork
4.0
6.85
-22
.135
.510
,000
7,40
010
,000
7,40
019
8251
.9Pen
nsy
lvan
ia2.
82.
8-
21.0
29.8
300
300
62.2
Tex
asno
stat
ein
com
eta
x19
.026
.9no
lim
it15
,000
no
lim
it30
,000
61.6
The
tax
brac
kets
are
thos
eap
plic
able
in19
98,w
hile
the
exem
ptio
nis
the
inco
me
exem
ptfr
omta
xati
onfo
rsi
ngle
filer
s.T
heC
alifo
rnia
exem
ptam
ount
refe
rsto
ata
xcr
edit
.Tax
data
isco
nstr
ucte
dus
ing
inco
me
from
the
Mar
chsu
pple
men
tof
the
CP
Sfo
r19
80-1
999,
and
usin
gta
xes
repo
rted
from
the
NB
ER
TA
XSI
Mpr
ogra
mm
e.‘M
argi
nalta
xra
te’re
fers
toth
em
ean
mar
gina
lta
xra
teac
ross
hous
ehol
ds,
the
‘tax
brac
ket’
isth
em
ean
tax
brac
ket
acro
ssho
useh
olds
whi
le‘in
com
eco
mpr
essi
on’r
efer
sto
1m
inus
toth
era
tio
ofth
est
anda
rdde
viat
ion
ofin
com
ebe
fore
taxe
sto
the
stan
dard
devi
atio
n
ofin
com
eaf
ter
taxe
s(a
ndtr
ansf
ers)
.T
hein
com
eco
mpr
essi
onm
easu
reac
coun
tsfo
rbo
thst
ate
and
fede
ralta
xes.
The
bank
rupt
cyex
empt
ions
are
thos
eap
plic
able
tosi
ngle
filer
son
1Ja
nuar
yin
1984
and
1998
,w
hile
‘oth
er’re
fers
toth
em
oney
amou
nton
allas
sets
excl
udin
gho
usin
gan
d‘t
ools
of
trad
e’.
Cal
iforn
iare
fers
tosy
stem
Iex
empt
ions
.T
heco
lum
n‘fe
d’re
fers
toth
eye
arin
whi
chth
efe
dera
lex
empt
ion
was
not
allo
wed
,w
hile
‘deb
tors
’
refe
rsto
the
prop
orti
onof
hous
ehol
dsin
the
stat
ew
ith
atle
ast
som
eun
secu
red
debt
.
48
Table 6: The effect of taxes and bankruptcy exemptions on unsecured debt
Income Mean marginalcompression tax rate
tax 0.381 0.442(0.903) (2.523)
exemption × (1-house) -0.028 -0.048(0.284) (0.305)
exemption × house 0.213 0.214(0.054) (0.061)
house fully exempt 0.576 0.559(0.190) (0.178)
age/10 -6.404 -6.565(3.231) (3.002)
age/10-squared 1.616 1.651(0.753) (0.709)
age/10-cubed -0.136 -0.139(0.057) (0.054)
finished school 0.956 0.943(0.153) (0.149)
some college 1.259 1.244(0.153) (0.154)
full college degree 0.761 0.739(0.158) (0.154)
black -0.574 -0.577(0.126) (0.116)
female -0.208 -0.210(0.068) (0.062)
couple -0.228 -0.219(0.222) (0.236)
ln(family-size) 1.092 1.076(0.379) (0.400)
ln(family-size)-squared -0.404 -0.396(0.168) (0.173)
ln(income) 27.153 27.033(2.943) (3.266)
ln(income)-squared -1.287 -1.281(0.141) (0.157)
interest rate -0.062 -0.055(0.323) (0.296)
Estimated by Censored Least Absolute Deviation (CLAD), with bootstrapped
standard errors, in parentheses, using 100 repetitions. Regression included
all households in the 18 largest states whose head was between 30 and 60
years old. Month, year and state dummies are included. Unsecured debts and
the bankruptcy exemptions are measured in logs. ‘House fully exempt’ is a
dummy with the value one if there is no upper limit to the value of the housing
exemption. The interest rate is the real municipal bond rate. The sample size
is 34,085.
Tab
le7:
The
effec
tof
taxes
and
ban
kru
ptc
yex
empti
ons
onco
nsu
mpti
onin
sura
nce
inco
me
com
pre
ssio
nm
ean
mar
ginal
tax
rate
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
sd(c
it)
sd(∆
c it)
sd(∆
c it)
sd(∆
c it)
sd(c
it)
sd(∆
c it)
sd(∆
c it)
sd(∆
c it)
tax
-0.0
76-0
.254
-0.9
61-0
.743
-0.1
88-0
.485
-0.5
58-0
.541
(0.0
50)
(0.0
83)
(0.4
95)
(0.2
29)
(0.0
70)
(0.1
17)
(0.1
38)
(0.1
35)
exem
pti
on-0
.047
-0.0
66-0
.093
-0.0
55-0
.042
-0.0
52-0
.050
-0.0
49(0
.014
)(0
.024
)(0
.033
)(0
.034
)(0
.014
)(0
.023
)(0
.023
)(0
.034
)hou
sefu
lly
exem
pt
-0.1
48-0
.108
-0.1
18-0
.014
-0.1
22-0
.046
-0.0
12-0
.027
(0.0
51)
(0.0
84)
(0.0
96)
(0.1
27)
(0.0
51)
(0.0
85)
(0.0
81)
(0.1
30)
const
ant
0.85
60.
719
0.94
50.
729
0.83
90.
665
0.64
90.
634
(0.0
67)
(0.1
13)
(0.2
18)
(0.1
79)
(0.0
66)
(0.0
54)
(0.0
55)
(0.0
39)
IV
lag
pol
lag
pol
Rank−
test
5.45
6.94
79.7
55.7
(pro
b)(0
.000
)(0
.000
)(0
.000
)(0
.000
)Sarg
an
10.7
78.
91(p
rob)
(0.0
56)
(0.1
12)
N42
041
238
435
842
041
238
435
8R
20.
143
0.09
10.
153
0.10
8
Stan
dard
erro
rsin
pare
nthe
ses.
IV
refe
rsto
whe
ther
the
tax
syst
emis
inst
rum
ente
dby
itse
lfla
gged
twic
e
(den
oted
‘lag’
)or
bya
set
ofpo
litic
alin
stru
men
ts(d
enot
ed‘p
ol’)
.A
llre
gres
sion
sin
clud
eda
full
set
ofst
ate
dum
mie
s.T
heco
lum
nsar
eov
erla
bele
dw
ith
the
mea
sure
ofth
eta
xsy
stem
that
was
used
inth
ere
gres
sion
.
50
Table 8: The relationship between taxes and bankruptcy exemptions
income compression mean marginal tax rate
(1) (2) (3) (4) (5) (6) (7) (8)
OLS IV IV IV OLS IV IV IV
tax -0.049 -0.316 -0.269 -0.234 -0.171 -0.236 -0.201 -0.174(0.016) (0.119) (0.071) (0.034) (0.020) (0.028) (0.029) (0.013)
constant 0.069 0.161 0.139 0.128 0.093 0.109 0.099 0.092(0.007) (0.040) (0.024) (0.012) (0.006) (0.007) (0.007) (0.004)
IV lag pol1 pol2 lag pol1 pol2Rank − test 4.98 6.94 7.16 55.7(prob) (0.000) (0.000) (0.000) (0.000)Sargan 2.965 42.78 4.290 27.99(prob) (0.085) (0.000) (0.038) (0.000)N 420 384 358 358 420 384 358 358R2 0.746 0.777
Standard errors in parentheses. IV refers to whether the tax system is instrumented by itself
lagged twice (denoted ‘lag’) or by a set of political instruments: ‘pol1’ denotes the regression
in which we use only the political affiliation of the state legislature and the tax efficiency
index as instruments, while ‘pol2’ denotes the regression in which the full set of instruments
is used. All regressions included a set of state dummies, and a dummy for Minnesota post
1993. The columns are overlabeled with the measure of the tax system that was used in the
regression.
51